WRD 2020 ISRRT SPECIAL EDITION CONTENTS Message from the ISRRT President 3 Thematic approach – practicing radiographer’s view: Claus Brix, Denmark 4 Ramona Chanderballi, Guyana 6 Edward Chen, Hong Kong 8 Håkon Hjemly, Norway 10 Naoki Kodama, Japan 12 Angela Meadows, United Kingdom 14 Yudthaphon Vichianin, Thailand 17 Denise Choong, Singapore 20 Jo Page, Australia 22 Alexander Peck, United Kingdom 24 Noeska Smit, Netherlands 26 Jill Schultz, USA 28 Christopher Steelman, USA 30
ISRRT Societies – member views and initiatives on Artificial Intelligence: ASRT 34 Cambodia 35 Ireland 36 Japan 38 Nigeria 39 South Africa 41 Singapore 42 Trinidad & Tobago 43 Tunisia 48 United Kingdom 51
Contributing stakeholders views: Adrian Brady, ESR 54 Nick Woznitza, Spencer Goodman, Jonathan McNulty, EFRS 56
In this Special Edition, articles and reports by authors/societies do not necessarily reflect or represent the opinion and thesis of the ISRRT.
2 Message from the ISRRT President
By Donna Newman
Dear ISRRT Members and Radiography image quality contributing to the efficiency radiological technologist elevating patient Professional Stakeholders, and increase in patient safety. care with Artificial Intelligence.
WORLD Radiography Day will be celebrated AI technology is also allowing for dual energy Additional with our members and health on November 8,2020 around the world by detectors to image both soft tissue and bone professionals around the world we hope that our profession. density in one imaging procedure which also by reading this edition and incorporating allows improvement in image quality while what you learn from our authors you will As we celebrate it is good to know that we reducing patient dose. help drive the ISRRT key messages of all have played a vital role in the delivery of creating, influencing and impacting change medical imaging and radiation therapy for AI tools in sonography and echocardiography within your country and daily practice. As patients all over the world. I’m thrilled to are aiding in mapping patient anatomy, AI technology advances radiographers will share this second special addition of World improving reproducibility which will focus on what they do best, to act as an Radiography Day “Elevating Patient Care improve image quality. AI systems have interface between the patient and imaging with Artificial Intelligence: Radiographers helped achieve accurate measurement of and radiation therapy procedures and use are essential in elevating patient care with Left Ventricular Ejection Fraction (LVEF), their compassionate human touch to care artificial Intelligence”. The theme which was which has contributed to consistent and for the patient during those difficult times chosen to raise awareness, help educate accurate results. for patients which is essential for the and shape the perception of the vital role delivery of an efficient, effective and caring radiographers/ radiologic technologists play AI tools in radiation therapy have brought health care system. as part of the health care team in elevating about improved access to the patient’s patient care while using artificial Intelligence diagnostic imaging, creation of AI contours, during our work day. dose calculation for the scheduled plan as well as AI driven adapted treatment plan, Today, we see Artificial Intelligence (AI) evaluate, check plan and deliver. providing useful tools for the radiographer’s ISRRT has collaborated with ISRRT member daily practice to enhance workflow, including experts, ISRRT member societies and worklist designations, workflow distributions regional stakeholders to highlight some of and scheduling assistance relating to these AI technological advancements used reducing no shows, appointments and repeat when performing medical imaging and examination. Radiation Therapy procedures.
Also available today is AI software that can ISRRT hopes that this educational material help to automate Quality control, lower will demonstrate and highlight how radiation dose, improve positioning and radiographers, as experts, use their critical improve sequencing in imaging. thinking and judgement when evaluating positioning and determining if adjustments AI is being used in specialties such as MRI, are needed for software settings when CT, Nuclear Medicine, Hybrid Imaging and incorporating these AI tools. Molecular Imaging such as PET CT and PET MRI to improve image acquisition and Additionally, ISRRT hopes this addition sequencing as well as enabling radiation captures the essential aspects dose reduction, reduction in imaging time that radiographers play in patient and enhancing postprocessing. communication, patient assessment, patient monitoring and most importantly Machine learning software has been the human touch during patient care during developed to help position patients in the a procedure. iso center when performing scanning which help ensure faster speed and precision while We hope you enjoy this publication and that ensuring use of optimal exposure control. you will find relevant resources, educational These tools can help lower dose and improve tips and ideas that will help radiographers /
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 3 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW An insight into Danish radiography and the work with Artificial Intelligence (AI)
By Claus Brix, Director of Professional Policy, Danish Council of Radiographers
THE Danish Healthcare System is one of the improve our organizations and start working most developed in the world and with free towards identifying areas where AI can help hospital service to all citizens, we welcome us create the necessary time and prevent AI as the technology which can make the radiographers and other professionals in best use of resources for the benefit of the radiology departments get ill from stress patient, the diagnostics and treatment. related symptoms. Pica Anderson Danish Council of Radiographers has asked Chief Radiographer, two radiology departments in different So where and how can we start? We can for Kolding Hospital, part of Denmark to give their vision about instance start by identifying which time- Sygehus Lillebaelt, Denmark AI in the future of radiology with focus on consuming tasks we perform today and challenges and benefits. continue by identifying which of these tasks new technologies/AI can manage. Here are a few examples of tasks we spend Aspects of Artificial Intelligence (AI) time doing and instead we can use this time in radiography to create excellent patient care: Pica Andersen, • We spend time positioning the patients Chief Radiographer, correctly in CT scanners. If the scanners Kolding Hospital, can be built to identify the correct Sygehus Lillebaelt, Denmark iso-center for the patient, the correct Helle Precht, start and finish for the scout image and PhD, University of Southern Denmark/ corrections for any abnormalities, the Kolding Hospital, radiographers can concentrate on the Sygehus Lillebaelt, Denmark patient. Helle Precht PhD, • We spend time finding blood test results University of Southern Denmark/ New technologies are moving forward prior to patient examinations. If an AI Kolding Hospital, and we, Radiographers need to keep our system can identify which blood tests Sygehus Lillebaelt, Denmark focus on how these inventions can we need for a given examination, send increase patient care or help us increase a notification to the patient and the patient care. laboratory and place the results in the RIS system, the radiographer and the As we speak, the population of citizens booking personal can save time. above the age of 80 is growing and by 2057, • We spend time creating reconstructions estimated one in ten citizens in Denmark of CT scans. This should be an easy task will be 80 years + compared to today where for AI. 4.4 % of the population are above 80 years The radiologists and radiographers old. Currently the 4.4 % of 80 + years old’s spend time measuring vessels prior to use 10 % of the local GP’s capacity so this interventional radiology. If AI can measure means we need to prepare our healthcare vessels from the CT/MRI scans and identify system to embrace the increased demands the equipment we need for the operation, Jakob W. Poulsen the future holds. In radiology, this will this can free a lot of time. Chief Radiographer, result in more scans and increased Bispebjerg and Frederiksberg demand for interventional radiology, more Quality assurance is another area where AI Hospital, requests for mobile radiology such as can help us create a safer environment for Copenhagen, Denmark bedside ultrasound, X-rays performed in our profession and our patients: nursing homes and other time-consuming • Help guide to correct positioning when patient related tasks. With no increase in taking X-ray images. government funding, we must find ways to • Help guide for the exact correct amount
4 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 of mA, time and kV for each patient needed. But it will also help radiographers examination according to the potential The development and use of Artificial to be able to concentrate more efficient on diagnosis asked for in the referral, Intelligence (AI) is growing rapidly these the patient care itself, and thereby to focus meaning another image quality is years and it is, without any doubt, here to even more on the humanity part of our job. required for scoliosis than for chest stay. In the following, I will try to describe, In my point of view, AI will have the following examination or between a first time how we work with AI at my department, value and effect: image and a control image of a fracture. and how we think it will elevate the Patient • Automatic identification of right and left Care in the years to come. By Patient care It will promote the – quality, efficiency, according to the patients anatomy, and I mean the quality of examinations, images patient safety, speed and precision. prevent wrong side markers. and reports but also the possibilities that the • Let the system report if the dose product technology gives radiographers to increase It will reduce the – lack of radiologists, for an examination changes according to the quality of the specific way we treat and radiation doses, use of contrast agents, the normal used protocols. take care of patients. image noise, number of retakes and errors, costs and moral injury (burnout) of staff. Also, in the booking process, AI can As we all know, the imaging options are contribute: increasing, and to help us all manage As radiographers in the ongoing transition, • Ensure that patients needing reoccurring the increasing numbers of examinations we must not forget the humanity part of our control scans get booked in the same and images, not to mention the lack of job and we need to secure, that we are up to scanner or x-ray room. radiologists, we need to find areas, where AI speed with the upcoming technologies. Let • Book certain patient studies such as can help us save more time and, at the same us embrace AI. children, dementia patients or other time, increase quality of all we do. specialized patient categories on days In the Capital Region of Denmark, where I And on top of the comments above, I have where specially trained radiographers work, there are a lot of initiatives to develop not even mentioned the arrival of robots are at work. and implement AI directly into clinical in radiography/radiology. Who knows – we • Ensure that patients needing reoccurring practice. Many of these initiatives are might even end up with a self-service CT- imaging are booked to meet the same performed as private and public cooperation scanner? radiographer each time they come. and many of the hospitals and radiology • Overview and note on the screen for departments in the region is involved in the the radiographer if the patient has process. These initiatives are performed in experience with the specific examination the areas of neuro-, abdominal-, thoracic- referred to, to optimize the radiographers and musculoskeletal radiology. communication and possibilities to help the patient thought he examination. At my department, almost all of the work related to AI is handled by our Professor and We also need to think ethics into areas his staff. He is a very broadly embracing and where AI can contribute and remember innovative radiologist, and therefore a lot of to think the patient experience into the the work is handled as a fusion of the classic equation. An example can be software way of research and modern innovation. The in MRI that helps decide the amount of benefit for our patients is obvious. We seek frequencies required for correct diagnosis. to implement AI on the fly and as soon as If the software detects any pathology in we trust the results. At the moment, most the first frequency it will decide for more of our projects is performed in the area of frequencies, but if it finds no pathology the musculoskeletal radiology and we work scan will end. This software will save time closely with private partners and developers. but can potentially give the patient an idea The professionalism of radiography is of the result according to the length of time changing rapidly and the transition from 2D spent in the scanner and cause the patient to 3D is going to change our profession even anxiety. more, I think. We will still perform a certain amount of conventional images, but I think we will see a huge increase in 3D-images in Elevating patient care with both the musculoskeletal area as well as in Artificial Intelligence the thoracic area. In this transition, AI will Jakob W. Poulsen, have a big impact. And we must not fear this Chief Radiographer transition – we need to embrace it! at Bispebjerg and Frederiksberg Hospital, Copenhagen, Denmark AI will help us all optimize workflows and precision in the outcome. It will help Radiography is changing and it is changing radiologists so they can spend their time very fast! more wisely and put their efforts where it is
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 5 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW Does the future of AI in medical imaging have a home in the developing world?
By Ramona Chanderballi, Guyana
Introduction echocardiography in the identification and WORLD Radiography Day 2020 is themed management of heart disease, the leading “Elevating patient care with AI; cause of death in most developing nations, Radiographers are essential in evaluating mechanisms to improve access to cardiac patient care with AI”. Artificial intelligence, diagnostics must be sought in order to Ramona Chanderballi simply put, is computer programming improve patient care and outcomes. Ramona entered the medical learning to think like humans. While this imaging field with her Certificate may be the new cutting-edge technology How can AI help? in Radiography in 2010 from the in the radiology world, does AI in medical In 2019, FDA gave clearance for the Ministry of Health, Guyana. imaging have a home in the developing echocardiography image analysis system world? “EchoGo Core” for commercial use by She was employed as an X-Ray clinicians. Some of EchoGo capabilities Technician at the Georgetown The earliest known publication of an AI are for automated performance of Public Hospital Corporation, article was in the 1950s, about 60 years parameters such as Left Ventricular Guyana from 2010 to 2014 when since the discovery of x-rays by Wilhelm Ejection Fraction (LVEF) and global she left to complete her BSc in Roentgen. And with the massive, rapid longitudinal cardiac strain, among others. Medical Imaging at the University advancements the medical imaging field At this point, cardiac strain analysis is not of Guyana where she graduated in has seen in the last few decades, it was routinely utilized in many centres and is April 2018. only a matter of time before AI became a not available in Guyana. However, LVEF major player in the technology used in the is routinely evaluated, and is an essential She is the owner and blogger radiology. MRI, CT, plain film radiography, component of each echocardiographic of RadToTheBone592, the first hybrid and molecular imaging have study. Accurate measurement of LVEF, like medical imaging blog in Guyana. all successfully employed AI in image many other aspects of echocardiographic Ms Chanderballi is currently interpretation,echocardiography has been study is subject to operator experience a cardiac sonographer at the catching up in the last few years. and diagnostically accurate image Georgetown Public Hospital quality. The use of AI systems can reduce Corporation and is the first female Echocardiography in the developing world user variability in both calculation and cardiac sonographer. She served Guyana is a country with a population of interpretation, leading to more consistent as the WRETF Ambassador for about 800,000 and is classified as a low and accurate results. Guyana July 2018 to July 2019, to middle income developing country. and currently sits as the youngest There are currently only five American According to a study published in 2018 Trustee on the WRETF Board where Society of Echocardiography (ASE) level by Alsharqi et al, from 2007 to 2018 AI she functions as the Ambassador trained cardiac sonographers performing was able to learn calculation of EF and Program Director. echocardiograms in Guyana. This longitudinal strain, quantification of wall translates to one cardiac sonographer motion abnormalities, quantification Ramona is currently in training for for every 160,000 persons. Of the five full of mitral regurgitation, classification/ pediatric echocardiography. time cardiac sonographers, only one is discrimination of pathological patterns trained in the performance of pediatric (Restrictive cardiomyopathy vs cheat echocardiography with a second cardiac pain, and Hypertrophic cardiomyopathy sonographer currently in pediatric training. vs Athlete’s heart in 2016), assessment of myocardial velocity, and recognition of 15 This small number of qualified cardiac echocardiography views. sonographers is inadequate to meet the demands for echocardiography in Role of AI in enhancing accuracy Guyana, and a similar lack of appropriately of echocardiography in a limited qualified and experienced sonographers resource setting exists in many, if not most, developing It takes on average of 10 years for a countries. Given the important role of physician to complete medical school and
6 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 specialty training to attain proficiency Although AI is costly and is currently in echocardiography. For cardiac exploring its uses in the developed world, DAIC. 2020. FDA Clears AI-Powered Cardiac sonographers, it can take anywhere from partnerships between low- and middle- Echo Analysis And Quantification By 2 to 4 years of university education, then at income countries and high-income Ultromics. [online] least 5 more years to become adequately countries are being studied with the hopes Available at: proficient. In resource limited countries, of broader reproducibility. But limitations www.dicardiology.com/content/fda- patients cannot wait years for services the exist in technological infrastructure, human clears-ai-powered-cardiac-echo- need now. expertise, and legislature. analysis-and-quantification-ultromics Mahajan, A., Vaidya, T., Gupta, A., Rane, The use of machine assisted learning in According to the International Development S. and Gupta, S., 2019. Artificial echocardiography was highlighted in a Innovation Alliance (IDIA), there are intelligence in healthcare in paper published by the American Society currently 4 billion people without internet developing nations: The beginning of Echocardiography titled “Artificial services. Until plans can be put in place to of a transformative journey. Cancer Intelligence and Echocardiography: ethically integrate AI in finding solutions in Research, Statistics, and Treatment, A Primer for Cardiac Sonographer”. a limited resource setting, AI will widen the 2(2), p.182. Algorithms were developed to give gap between the services is has available Narang, MD, FACC, A., 2019. Artificial guidance to new sonographers on and who can afford to access them. Intelligence and Echocardiography. technically correct image acquisition by [Blog] American College of Cardiology, the software recognizing incorrect or off AI, Echocardiography and the future Available at: axis views and providing real time guidance AI, while of significant potential benefit www.acc.org/latest-in-cardiology/ on how to move the probe to acquire the in the provision of cardiac diagnostics, articles/2019/06/18/07/43/artificial- correct image. can never replace the importance of intelligence-and-echocardiography the one-to-one interaction between the Sengupta, P. and Adjeroh, D., 2018. Will The paper also highlighted advantages cardiac sonographer and the patient. Artificial Intelligence Replace the of AI assistance in a high patient volume Echocardiography will always remain both Human Echocardiographer? Circulation, echocardiography lab. With the increased an art and a science. Utilizing the best 138(16), pp.1639-1642. need for echocardiographic studies in a available technology is critical to optimizing Wahl, B., Cossy-Gantner, A., Germann, population with longer life spans and body the performance and interpretation of S. and Schwalbe, N., 2018. Artificial mass indices, sonographers are at a echocardiographic studies by the cardiac intelligence (AI) and global health: how higher risk of musculoskeletal injuries. sonographer and the interpreting physician, can AI contribute to health in resource- Automated acquisition could possibly serve but will not negate the role of the human poor settings?. BMJ Global Health, 3(4), as a solution to improve the ergonomic art of medicine. p.e000798. burden sonographers face with increased patient volumes. References Alsharqi, M., Woodward, W., Mumith, J., AI implementation in the developing world Markham, D., Upton, R. and Leeson, Despite the significant costs of AI P., 2018. Artificial intelligence and implementation, there are several AI echocardiography. Echo Research and programmes being tested in various parts Practice, pp.R115-R125. of Africa and India in the health industry 2019. Artificial Intelligence In International with the potential for reproducibility in Development: A Discussion Paper. [PDF] other resource limited communities. Available at: static1.squarespace.com In an interview with SciDev.Net, Wadhwani Mobile for Development. 2020. Can Institute for Artificial Intelligence senior AI Help Tackle The Most Pressing programme director- Mr Neeraj Agrawal Challenges In Developing Countries? outlined 7 questions Wadhwani AI uses to [online] Available at: decide whether or not an AI project will www.gsma.com/mobilefordevelopment/ benefit a community. The questions he africa/can-ai-help-tackle-the-most- outlined are: Is this a big problem? Does pressing-challenges-in-developing- it have an AI solution? Will solving the AI countries/ part make enough of a difference? Will the [Accessed 1 August 2020]. solution be accepted by stakeholders? Does Davis, A., Billick, K., Horton, K., Jankowski, the data exist, or can it be created easily M., Knoll, P., Marshall, J., Paloma, A., enough? Are there partner organizations Palma, R. and Adams, D., 2020. Artificial that can co-create and pilot the solution? Intelligence and Echocardiography: Are there existing programmes and A Primer for Cardiac Sonographers. pathways to scale? Journal of the American Society of Echocardiography.
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 7 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW The radiographer is promoting and maintaining a safety culture
By Edward Chen, Hong Kong
Introduction – Information Management AI in Healthcare service is springing up like It is like the communication system of mushrooms in the last decade, especially the patient and healthcare professionals, in radiology services. An AI article in the such as the mentioned auto customer recent Healthcare Weekly has tried to relationship system. Some RIS vendors illustrate the renouncing AI products. can provide the AI self-booking system There are nine out of twelve products1 for the patients. Nowadays, many HIS can related to radiology or medical imaging. provide allergy or contra-indication alert However, all of the radiological AI products to the radiographer or nurse before the are supporting the diagnosis, serving the examination. Those tools can minimize Edward Chan radiologists. Does AI help the radiographers human error. Some radiography school has Edward achieved his basic or radiological technologists? And does installed the VR teaching system, which is radiography training in Hong it elevate the patient care? This article another format of AI tools of information Kong and Australia and has two is trying to organize the information of management. Master Degrees in Psychology healthcare AI to see whether radiographer and Health Care. He is the Vice- can apply the AI and our role of its Roles of Radiographer in AI application President of Hong Kong College development. I hope this article can As I mentioned, we have applied AI in of Radiographers and Radiation inspire you to have more new idea of AI medical imaging for some time. Many Therapists. From 1998 to 2012, development in radiography. healthcare professionals have joined the he was the Chairman of the Hong development and research of AI business. Kong Radiographers’ Association. Clinical application of AI Directly speaking, there are two groups He organised various functions The term AI was first used in 1956 by a of people in AI application, User and throughout the years, such as group of scholars and computer scientists Developer. continuing education programs of Dartmouth College. They made it to & assessments, newsletters, describe the machine which could solve – User seminars & conferences. problems as the human using natural Radiographers, as a user, provide efficient Promoting professional intelligence.2 After over 60 years of and accurate services by AI assistance, such development, liaising with local and development, AI is affecting our daily life as automatic 3D processing, lung nodule overseas relevant organisations even though you may not notice. It has detection and fracture detection.1 They are his missions throughout these been used in areas such as the auto- understand the operation of the AI system two decades. Providing expert braking system of your car and the auto and know how to interpret the results of AI. recommendations to Government customer relationship system either online The interpretation is the most important and related institutes about or on the phone. of AI application to deliver clinical service. medical imaging services are his Black box difficulty is a jargon of AI. That duties of the professional bodies. In the healthcare field, the author of means AI cannot ask “why” during their Standard of Practice and Radiation Healthcare Weekly, Cordrin Arsene, gave learning process. AI, either deep learning Protection is the focus of his career. a summary of the AI applications1 which or machine learning, is trying to find the Besides that, Edward has been the consist of two dimensions, Decision Support pattern of the datasets. The logic behind the Certified Professional Healthcare and Information Management. result is not well defined. Users’ expertise Information and Management becomes essential for interpreting System (CPHIMS) since 2008. He – Decision Support the decision and limiting the negative is one of the founders of medical It is like most of the AI tools in the radiology implication. imaging informatics profession department. They are helping in Hong Kong. Recently, AI the radiologist to find out the lesion from – Developer development of radiography the images by faster and more accurate Radiographers can be an AI developer becomes his new studying topic. pattern recognition and providing the by joining AI researches and providing clinical suggestion, such as the most professional opinions to develop AI email: [email protected] common lung nodules or lesion recognition algorithms. Helping to collect meaningful of virtual colonoscopy of CT. Those tools data for AI learning is essential to build up are trying to do early detection and improve the AI system. Radiographers can take this efficiency. role well too.
8 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 Examples of elevating patient care the AI to learn the contrast enhancement Exposure Control). Those are helping us With the applications and roles, I made a pattern. Indeed, the professional who gives to speed up the process and reduce the diagram to illustrate the examples of usage the injection has to decide trust or modify error. AI is serving the same. I want to and position of radiographers that they the protocol in practice. quote a statement of Dr Nick Woznitza, can take part in AI. The information on it is co-chair of EFRS-ISRRT working group inexhaustive. You are welcome to add more – Allergy and contrast-indication Alert on AI. “Radiographers may act as in the future. Straightly speaking, it is not an AI system. the gatekeepers to medical imaging, using AI as a decision support tool to ensure the examination performed is correct.” he made it in an interview by ESR4. To embrace AI in radiography, we need a good foundation in our basic professional knowledge and the theory of AI development. Those are the elements to be a gatekeeper either in the application or development. The radiographers can refer “A Joint Statement of the International Society of Radiographers and Radiological Technologists and the European Federation of Radiographer Societies.5” It gave some cardinal directions about applying AI in our profession.
Reference 1. Arsene C, Artificial Intelligence in Healthcare: the future is amazing, Diagram 1: Roles and Applications of AI in radiography. Healthcare Weekly, July 2020 https:// healthcareweekly.com/artificial- It is a keyword searching system within the intelligence-in-healthcare/ – QA of images patient information. Many HIS systems have 2. Castelo M, The future of Artificial Some research groups have developed this build-in. However, the continuous input Intelligence in Healthcare, their AI systems to check the quality of of radiographers is necessary because the HealthTech Magaizne, 2020, https:// the radiographs. They can maintain the standards and requirements may change healthtechmagazine.net/article/2020/02/ quality of images for the radiologist to from time to time. The development of future-artificial-intelligence-healthcare report. Radiographers have to select a lot keywords dictionary must be a collaboration 3. Imaging Technology News, Using of standard radiographs for the system work among all healthcare professionals. I Artificial Intelligence to Reduce to learn. The AI system can give a rating did that for my hospital’s HIS. It was tedious, Gadolinium Contrast, 2020, www. and the false of the quality. In reality, the but the functions are useful for patient care. itnonline.com/article/using-artificial- operating radiographer has to decide intelligence-reduce-gadolinium-contrast whether to accept its comment or not. – Self-appointment booking system 4. ESR, What the increasing presence of AI Some HIS and RIS services providers have means for radiographers | AI Blog, 2019, – Personalize CT contrast injection protocol introduced this function in their systems. https://ai.myesr.org/healthcare/what- Some researchers have developed the AI For RIS, examination booking is complicated the-increasing-presence-of-ai-means- systems to learn the difference of full dose because the system has to consider the for-radiographers/ contrast MRI images and low dose contrast contrast indication, preparation, availability, 5. EFRS-ISRRT, Artificial Intelligence images3. Then the system can reconstruct etc. AI should gather all the information and the Radiographer / Radiological the low dose images as the full dose quality. to provide a recommend time slot for the Technologist Profession, A Joint My example is a hypothetical system. As patient to accept or allow the patient to Statement of the International Society we all know, it is a challenge to achieve choose a possible one. On the other hand, of Radiographers and Radiological the best contrast enhance result during the system should provide all the necessary Technologists and the European the CT contrast exam even though we have information for them to know and follow. Federation of Radiographer Societies, tons of researches studying about it and 2020 generated various protocols. Bodyweight, Summary age, heart rate and BP can affect the AI is a concept for a long time. With the contrast bolus flow and enhancement of advancement of computing technology, the organs. With AI support, radiographer it is becoming part of our daily life. The could have a personalized injection protocol motivation for developing AI is similar to achieve the best enhancement. So, we to create tools, such as CAD (Computer need datasets of different patient types for Aided Diagnosis) tools and AEC (Automatic
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 9 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW Radiographers and AI, do we have to change our ethics?
By Håkon Hjemly, Norway
Introduction service are critical, it is necessary that they A PROFESSION is often recognised by its are rely on the trust between the players. own set of ethical guidelines. These ethical Patients have the right to co-determination guidelines may be at both national and regarding decisions about themselves international level. This is also the case for with the principle of informed consent radiographers/radiological technologists important in most ethical guidelines for (radiographers), as many national societies radiographers. This means in practise that have developed their own in addition to the radiographers should be able to educate international guidance by the ISRRT1 or and provide appropriate information to other regional recommendations. patients and/or family, so that they can Håkon Hjemly understand and make informed decisions Håkon Hjemly has been the ISRRT Professional ethics delineates how about their care. Also, if an AI system Vice President Europe Africa broader ethical standards, such as acts in a way that we do not anticipate or since 2018. He is Manager of responsibility, integrity, fairness, understand, claiming ignorance cannot Policy at the Norwegian Society transparency and avoidance of harm absolve professional’s ethical responsibility of Radiographers. His main apply to the basis of the profession. This for the outcome. responsibilities for the society includes such elements as safe, caring are related to professional role and a knowledge-based application of I strongly believe that this ethical principle development of radiographers and advanced technological equipment in the is important to have in mind when new to health political issues. best interests of the patient, and that the systems based on AI are introduced. It is radiographer`s interaction with the patient essential therefore that the patients have He has a Masters Degree in shall be based on respect for human rights, full confidence in the radiographers and clinical health, focusing on role equality and justice. the tools they use, no matter how advanced development for radiographers, they are. and post graduate education in Radiographers are the interface between both digital imaging processing medical imaging- and radiotherapy What do radiographers need to know? and in x-ray protection. Prior to the technology and patients. Patients rely on Accountability for personal actions and for work for his Society, he had variety and trust the radiographer to make correct maintaining professional conduct is always of roles in both the private and decisions. Radiographers often need to relevant, but how far can accountability for public sector in Norway; Clinical explain to their patients what is planned, the actions of an AI system be taken? Radiographer, QA-administrator, why and what will be the next step in their AI technologies are not easy to understand Manager, Radiation Protection diagnostic or therapeutic pathway. The for those not involved in their design Officer, Product Specialist and Sales rapid development of software tools, with and development, and even those with Representative (CT, Mammography so called artificial intelligence (AI) and considerable knowledge on the subject and C-arm). machine learning (ML) features, challenges often find it difficult to understand how these ethical principles in different ways. software and devices with AI-content Håkon is also past president of the Medical imaging exams and radiation produce their output as their algorithms EFRS and served on the board for therapy procedures will most likely be remain opaque. They rarely provide any two terms. He is chairing on behalf done at higher efficacy and speed, involving human understandable explanation or of the ISRRT a joint project with the fewer professionals and include increased justification for their predictions. This EFRS focusing on radiographers throughput of patients. Ethics depends makes explaining the underlying logic of a role with artificial intelligence. on cultural values and changes over particular system very difficult, as to how time and my question is whether current decisions were made, whether there are professional ethics needs to change or be errors and how these might have occurred. looked at differently because of concerns One of the ways many have proposed to with the accountability and trust in address this issue is through ensuring autonomous systems with respect to AI? that how autonomous systems operate must be transparent to all the relevant Matter of trust and the principle of stakeholders. informed consent Because many choices made in radiology The term transparency also addresses
10 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 the concepts of traceability, explicability, checklists of how to be ethical in any of the current and future workforce and and interpretability. Transparency is also given situation a radiographer might the active engagement of radiographers essential for consent; no one can consent encounter, they are tools intended to and radiologic technologists in AI meaningfully when they do not understand help radiographers learn to judge what is advancements going forwards.” the implications of consenting. ‘appropriate’. But what if we experience that living up to our recommended ethical References The important point here is that standards is becoming more and more 1. www.isrrt.org/code-ethics responsibility cannot only be in the hands challenging? For example, if we often find 2. www.radiographyonline.com/article/ of the designer or vendor, but it must ourselves not able to understand or explain S1078-8174(20)30037-7/pdf be distributed across the process and why our AI-powered systems act as they its stakeholders. This requires that all do? Would it not be better to just accept stakeholders build awareness of possible this and to adjust our ethical standards issues, potentially including the use of risk accordingly? assessment and verification tools. This is also what I questioned in the The ISRRT and the EFRS jointly states2 headline and in the beginning of this that it is essential that radiographers and article; Do we need to change our ethical radiological technologists: standards or look differently at them with - Ensure that all research involving AI respect to AI? systems is conducted in an ethical way, communicating with patients on how I think we don’t. their data may be used to develop and test AI. I don’t recommend our ethical standards - Are involved in the piloting and to be lowered because AI is difficult to research of algorithms prior to clinical understand and explain. Lowering our implementation. standards when facing challenges is a - Understand how algorithms arrive at dangerous pathway to choose and could decisions and probability errors within lead to the profession being more and these decisions to enable effective more obsolete. I think how we adapt to communication of findings to patients. this game changing technology very much - Maintain and develop core skills and decides the future for the profession. Our competencies to act as a sense check patients and ourselves may trust such to AI-supported clinical decisions (for technology close to 100% in 10 years from example scan planning in CT, image now, but it is critical that we continue interpretation triage). keep our professional standards high if we want to be in control of it. I believe we This requires that radiographers need to will see radiographers working with more know and be able to explain, at least in responsibility and more independent from principle how AI based solutions delivers other professions with support of AI if we results with a high degree of accuracy follow the recommendations in the joint and precision. It also requires confidence ISRRT-EFRS Position Statement(2), that that there is control over the quality of the radiographers should adapt to this new data which AI methods and models are technology with our professional standards based on, and not least which tests and in mind and make sure we together elevate measurements the programme is subject patient care. to in the validation process. Radiographers must be able to supervise and evaluate “It is of critical importance that decisions and results that AI-tools radiographers and radiological produces and to override these if required technologists, as medical imaging and radiotherapy experts, must play an Need for lowering our ethical standards? active role in the planning, development, Radiographers are professionals; being implementation, use and validation of part of a moral community of others who AI applications in medical imaging and share the same responsibilities and being radiation therapy, reinforcing the need for able to draw on the experience of others the technology to be targeted to the most to navigate similar moral dilemmas, pressing clinical problems. The optimal tough decisions or adverse consequences. integration of AI into medical safety, clinical Although professional guides and codes imaging and radiation therapy can only be of conduct are not meant to be exhaustive achieved through appropriate education
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 11 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW Future roles of radiological and technologists and medical Artificial Intelligence
By Naoki Kodama, Japan
IN recent years, owing to the remarkable technologists, a series of processes from development of information processing image capture to diagnosis is the most and communication technologies, the important use of medical AI. In particular, spread of electronic medical charts and the Japan is rapidly aging, with a corresponding digitization of medical-related information increase in the number of medical images are rapidly progressing in the medical requiring interpretation that is not matched field. Medical device manufacturers are by an increased number of radiologists. developing artificial intelligence (AI)-based Medical AI can directly reduce the workload Naoki Kodama systems to effectively use these collected by helping radiologists to read images; Dr Naoki Kodama is a Professor medical data to provide higher-quality moreover, it and may also prevent and of Department of Radiological medical care. In addition, as the early reduce re-imaging and patient exposure Technology, Faculty of Medical detection of diseases is expected to reduce dose. In addition, medical AI can be used Technology, Niigata University medical costs, the Japanese government is to educate radiological technologists. It is of Health and Welfare in Niigata supporting the introduction of AI technology possible to learn normal and typical medical city, Japan. Prior to this he in the medical field (medical AI). findings, distinguish imaging findings worked in Department of (panic images) that require an emergency Medical Informatics, Faculty of While AI means “artificial intelligence”, it response, and improve image analysis and Health and Welfare, Takasaki is a program that reproduces the logical evaluation. Furthermore, it is possible to University of Health and Welfare, thinking performed by humans. Until now, improve the imaging technology to provide Japan, 2004-2017. He received it was a very limited intelligence that could higher quality medical images, reduce his bachelor’s degree in Medical only give answers under certain rules; patient exposure dose due to the difference Science from Suzuka University however, with advanced development of in imaging procedures, and perform more of Medical Science in 1999. deep learning technology since around effective dose measurement and dose He holds a master’s degree 2010, the programs learn from given data to evaluation in cancer treatment. in Engineering from Nagaoka build criteria without human input. Medical We introduce the two latest medical AIs University of Technology in 2001. care requires attention to disease risk that support the work of radiological He received his doctoral degree assessment, disease diagnosis, treatment technologists in Japan. in Engineering from Institute selection, and prognosis assessment. of Biomedical Engineering, However, these factors may be difficult to Detection support Nagaoka University of evaluate due to differences in individual (bone suppression processing) Technology in 2004. situations. The benefits of medical AI for Bone suppression processing generates Kodama holds professional both patients and medical staff, including an image from which the signals of the qualifications as a licensed radiological technologists, is that it is radiological technologist and possible to construct judgment criteria radiation protection supervisor. based on a large amount of accumulated Kodama has been the Director patient data and present the optimal of the Japan Association of treatment method for each individual. Radiological Technologists Medical treatment is roughly divided into (JART) since 2006, and vice- three steps; namely, prevention of disease president of the Japan Society onset, diagnosis to distinguish people who of Education for Radiological already have the disease, and treatment Technologists since 2008. In to improve the prognosis of patients with a 2018, he was elected as the diagnosis. Medical AI can contribute to any ISRRT Regional Coordinator of these three steps and is expected to be for Professional Practice (Asia/ utilized in all medical fields, following its Australasia). In 2020, Kodama increasing application in medical systems was elected Vice Chairman of including medical insurance systems. the JART. In medical images handled by radiological Figure1: Chest X-ray image (original image).
12 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 radiologists or radiological technologists. Furthermore, the method also improves the image capturing ability of radiological technologists and provides high-quality images that can contribute to early TB diagnosis. In addition, identification of abnormal findings at the time of imaging by radiological technologists may allow them to suggest additional imaging to doctors or examinations with another modality while the patient is at the hospital without returning home. This method is also expected to allow diagnosis and treatment.
AI to improve the examination workflow for radiological technologists Figure 2: Bone suppression image (BSI). The increasing awareness of medical Figure 4: Example of imaging failure safety underscores the necessity for (external rotation) and the judgment area clavicle and ribs, which are obstacles in exposure dose management systems that (pink). the interpretation of chest X-ray images, can centrally manage X-ray irradiation have been removed. Figures 1 and 2 show information such as exposure dose and automatically detects these various types a chest radiograph and bone suppression imaging protocols for individual patients of imaging failure is expected to provide image (BSI), respectively. Unlike energy base on X-ray imaging, computed educational support to reduce imaging subtraction, the feature is that the image of tomography (CT), and other diagnostic failure and improve technology. The use the bone signal is removed from the single imaging devices. By linking with the of this AI software allows the automatic plain X-ray image of the chest taken by a radiation work management support identification of the images that may normal system using the AI image analysis. system, various radiation inspection be missed; moreover, by visualizing the The doctors who interpret the images can data can be visualized to support both number of times, it is possible to identify reduce the possibility of missing the lesion the analysis and optimization of medical which technique results in the most hidden in the bone by alternately checking exposure. In addition to modalities with high missed images for individual radiological the original and BSI. In the bone recognition exposure such as CT, to reduce exposure technologists. It is also possible to carry process, the estimated value obtained doses due to general radiography with out targeted education. Furthermore, by from bone structure data collected from the highest frequency of radiography, creating a graph that is classified according significant chest X-ray image data and we developed AI-based software to to the cause of re-imaging combined with the estimated result of the bone structure automatically detect imaging failure needed an environment in which the cause of measured from the target chest X-ray re-imaging. These applications include the the loss can be analyzed along with the image are used to extract the detailed automatic detection of lung field defects failed image, it is possible to continuously structure. In bone signal attenuation, the (Fig. 3) and blurring in chest X-ray images, examine issues, study measures to prevent signal component of the bone is estimated which are frequently photographed, and loss, improve activities, and improve from the recognized bone candidate to AI, which automatically detects poor effectiveness. It can be expected to support perform attenuation. In this process, positioning of the knee joint (Fig. 4), which a measure to reduce re-imaging. As a only signal changes due to the bone are may cause image defects. Software that result, the patient burden is reduced by accurately estimated and attenuated, shortening the imaging time, the efficiency leaving the signals of fine structures such of imaging work is improved, and the as abnormal shadows and blood vessels inspection time for each imaging device and overlapping the bone, improving the lesion the operating rate of each imaging room visibility and reading efficiency. are visualized to allow the development of optimal placement plans for each device We performed receiver operating and radiological technologist. It will also characteristic (ROC) analysis on the support the comprehensive management of accuracy of tuberculosis (TB) diagnosis medical and radiation services. from BSI by nine Vietnamese radiologists using chest X-ray images of 325 cases. We observed significant differences between the three radiologists specializing in areas other than the chest. The area under the ROC curve (AUC) without and with BSI were 0.882 and 0.933, respectively. Figure 3: Example of the detection of a lung This method is very useful for educating field defect.
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 13 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW Elevating patient care in nuclear medicine & molecular imaging using Artificial Intelligence (AI)
By Angela Meadows, United Kingdom
Introduction for many years, and consider the benefits AS radiographers and radiological of AI in elevating patient care. technologists unite around the world on November 8, 2020 for World Radiography Background Day, there will be reflection upon a year What’s does the term Artificial Intelligence like no other. 2020 has been a pivotal (AI) mean? year globally, as we have had to adapt to a worldwide pandemic. We have had to “An area of study concerned with making adapt rapidly, and embrace communication computers copy intelligent human technologies we may not otherwise have behaviour”1 Angela Meadows been comfortable with. For many, video calling/conferencing is inconsequential As many AI definitions similarly suggest, AI Angela currently works with now, its second nature, but step back is the ability of a system to be able to copy Alliance Medical in the UK and is to 2019 and look forward and we could or simulate intelligent human behaviour responsible for managing the NHS not have imagined the way in which we or performance. However, there are two imaging service partnerships at communicate now, and how we have further key phrases to be aware of along the Preston PET-CT Centre and adapted as we socially distance as a means with AI, these are Machine Learning and the MRI service at Westmorland of global pandemic protection. Deep Learning (See Fig 1). General Hospital.
Similarly, when we look at the term Machine Learning is a subcategory of Angela has worked in imaging Artificial Intelligence (AI) we can conger up Artificial Intelligence that uses statistical service management for over 13 images from films like AI, and the Matrix. learning algorithms to build systems that years, managing PET CT and MRI/ We could imagine imaging technology have the ability to automatically learn and CT imaging partnerships in both which would require little human contact, improve from experiences without being the public and private sectors. machine learning technologies, computer explicitly programmed. We use these daily algorithms, data, data and more data. in search engines such as google and Angela’s specific interests are in yahoo or voice assistants such as ‘Siri’ PET CT, research, paediatric NM However, let’s stop and take a look at and ‘Alexa’.2 imaging and MRI safety. Nuclear Medicine (NM) & Molecular Imaging (MI), its advancements and AI Deep Learning however is a machine Angela is currently chair for the technology examples that have been in use learning technique that is inspired by the Nuclear Medicine and Molecular Imaging Advisory Group for the Society & College of Radiographers in the UK.
Figure 1.
14 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 way a human brain filters information, AI in NM & MI - Current & procedure takes place. Much research it is basically learning from examples. advancing technology is ongoing in the field of algorithmic It helps a computer model, to filter and Some of the first examples of AI in practice information extraction and what is input data through layers to predict and which remain in NM, involve the use also known as the data interoperability classify information. Subsequently, deep automatic organ segmentation techniques. standards such as Fast Healthcare learning processes information in a similar Computer based segmentation techniques Interoperability Resources (FHIR).3,4 Such manner to the human brain, it is the kind of allow Radiographers and RT’s to review and systems we hope will become gradually technology used in driverless vehicles.2 quantify image data presented as regions available in the clinical setting with of interest (ROI’s), this process provides concise case specific dashboards In NM & MI the very essence of our role as an analysis of regions and sub regions. becoming accessible with time. Improved a radiographer or RT is to work with ever Some of the first examples of quantitative access to data, means more timely approval developing complex imaging technologies, segmentation imaging are found in Tc99m for the imaging procedure and faster technologies which have evolved over DMSA renal imaging studies, which not imaging results are made available for the the years, which take complex computer only assess the size and location of the referring clinician. reconstructions and algorithms, which kidneys but more importantly the relative would typically have been manipulated renal function to determine if renal Following the scheduling of an NM or MI by the radiographer to now automate scarring may be present. In addition to imaging procedure, one of the necessary them for the radiographer, thus allowing Tc99m DMSA, more complex segmentation checks is to ensure all measures are the radiographer, to focus on the human techniques can be used in lung SPECT employed so that patients do not ‘Fail to patient care aspects of the imaging CT, to assist in the analysis of Pulmonary Attend’ (FTA) their imaging appointment examination, aspects such as physical and Embolism. Both of these techniques have without notice. This is particularly psychological wellbeing. When considering evolved over recent years and are good important in the field of NM & MI as patient care and management it is simpler examples of AI in practice, as in both tracer availability can be limited, is to consider the imaging procedure split into instances, automated analysis takes place subject to decay and there are of course three elements, preparation, during and using an applied computer algorithm which the associated costs and wasted vital post imaging procedure. When we consider automatically assumes the ROI for the appointments which could occur. Simple the imaging procedure broken down in this Radiographer, thus freeing up more time machine learning algorithms can be manner there are multiple points at which for the Radiographer to be with the patient employed to a database which can AI can feature to enhance the procedure post procedure. consider and review patient records for and assist radiographers & RT’s to ensure those which FTA. Using AI techniques of they can achieve the optimal image quality AI throughout the NM & MI this nature, which can consider imaging and patient care they strive for. imaging procedure procedure type, age and sociodemographic I previously explained that when we data, this may assist in pinpointing In NM and MI procedures such as PET- consider patient care and management it is trends for such FTA cases, allowing CT, there is an inherent complexity to simpler to consider the imaging procedure preemptive analysis and reminder image acquisition, from a static bone scan split into three elements, preparation, telephone calls or text message for the acquiring thousands of image counts during and post imaging procedure. Taking patient appointments.5 per second, to more complex hybrid the procedure as three elements we have technologies which integrate two separate examples of how AI can elevate patient AI in NM & MI - during the scanning systems such as PET with CT care. imaging procedure and SPECT with CT. Over the decades When considering AI in the NM & MI there are countless examples of where AI in NM & MI - Imaging preparation Imaging procedure, there are many computers have evolved with imaging Often a significant challenge in the different examples which could be technology advancements, to allow preparation and scheduling of an considered. Let’s take key examples for faster image processing, manage greater examination, is ensuring a concise yet SPECT CT and PET CT as complex hybrid volumes of data and present image detail appropriate medical history is detailed on technologies, these use AI to enhance and resolution better than ever before. We the imaging referral. If this information imaging capabilities and improve the have now moved into a time where imaging is not provided clearly by the referring patient experience. platforms present an intuitive approach for clinician, manually accessing this their users blurring the lines between the necessary information can prove difficult, Hybrid technologies use two independent radiographer and the technology to work particularly for imaging services which systems i.e. PET with CT or SPECT with as one. cover a large geographical area such as CT and integrate the images by image PET-CT, as referring hospitals do not fusion. These systems are processing In this article I wish to highlight some necessarily share the same databases for huge volumes of data and are making current examples of AI in NM & MI patient information. It is essential therefore use of machine learning (ML). Recent practice and consider how we work with AI that the Reporting Clinician is comfortable advancements in research suggest technologies to elevate patient care in this that the imaging procedure requested is considerable technical improvements in the field. appropriate to the patients clinical need, near future for such technologies.6 In NM thus ensuring no unnecessary radiation & PET CT, attenuation maps and scatter
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 15 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW correction remain a key discussion point for that the Radiographer, RT and Reporting professionals but we must ensure that for AI groups. Clinician all understand the strengths and optimum integration of AI into the clinical limitations of such algorithms and have imaging setting, appropriate education of I previously explained that AI now supports a sound knowledge of post procedure the current and future workforce and the an intuitive approach on the imaging image manipulation and its pitfalls. An active engagement of radiographers and platform, particularly when it comes to algorithm can only be as smart as the data RT’s in AI advancements is essential to day to day running of the systems. A good provided, so it is essential that this is fully ensure the safest and highest standards of example is the ongoing monitoring of understood by the operator. patient care.10 imaging systems in the background by the Original Engineering Manufacturer. When considering image reporting, References Often issues are flagged before they are predication and response to therapies are 1. www.oxfordlearnersdictionaries.com/ even apparent to the operator, as flags are areas where quantitative analysis can prove definition/american_english/artificial- raised to engineers, this allows them to a powerful tool if used correctly. Assuming intelligence arrange planned maintenance and address that NM and MI imaging scans performed 2. https://towardsdatascience.com/ potential technology issues which would pre and post treatment, are performed with understanding-the-difference-between- otherwise disrupt service without warning. consistency and accuracy, the resulting ai-ml-and-dl-cceb63252a6c Dose optimisation is also a significant focus image quantification using standardised 3. Pinto Dos Santos D, Baeßler B. across all NM & MI imaging, with again uptake values (SUV) will prove a reliable Big data, artificial intelligence, and hybrid systems using ML technology which tool to measure and make assessment for structured reporting. Eur Radiol Exp. is capable of personalising the scan field of treatment therapy response in oncology 2018;2:42. view based on anatomy. Such systems can cases. This will allow the referrer to 4. FHIR is a standard for health care data use single flowing patient scanning table confidently adapt treatment regimes for exchange, published by HL7®. movements to image patients top to toe in their patients based on imaging findings. www.hl7.org/fhir/ one flowing movement rather than section 5. Harvey HB, Liu C, Ai J, et al. Predicting by section image acquisitions, this enables Finally, when considering the generation no-shows in radiology using regression improved dose optimisation techniques and of the imaging report, this is also subject modeling of data available in the can often reduce the CT dose further. AI techniques. Again, techniques which electronic medical record. J Am Coll have been recognised for many years now, Radiol. 2017;14:1303–1309. As previously described, technologies in such as voice recognition which uses ML 6. Zhu B, Liu JZ, Cauley SF, Rosen BR, NM & MI use not only AI & ML technologies technology. This allows for prompt image Rosen MS. Image reconstruction by but also deep learning (DL). DL requires report generation and distribution direct domain-transform manifold learning. complex algorithms for more complicated to the imaging referrer if set up to do so. Nature. 2018;555:487–492 procedures such as simulation respiratory Slick image reporting techniques are of 7. BIO-TECH SYSTEMS, INC. Report 2008 gating for example. In 90% of PET-CT great importance in the imaging pathway 8. Grills, Inga S et al. “Potential for oncology cases, disease is located in the to ensure timely result generation to reduced toxicity and dose escalation in chest or abdomen.7 These areas are subject referring clinicians, which in turn ensures the treatment of inoperable non–small- to respiratory motion, which can displace prompt action and more immediate cell lung cancer: A comparison of organs and lesions by a range of 5mm patient care/interaction as a result of the intensity- modulated radiation therapy to 30mm, blurring images and reducing image findings. (IMRT), 3D conformal radiation, and diagnostic confidence.8 And without elective nodal irradiation.” International respiratory gating, 40% of lung lesions Conclusion Journal of Radiation Oncology • Biology may even go undetected.9 The use of DL This article has provided an overview • Physics, Volume 57, Issue 3, 875-890 technology in these cases means that no of the different terms used in AI. This 9. Garcia Vicente AM, et al. (18) additional scan time, count acquisitions includes the concept of Machine Learning F-FDG PET-CT respiratory gating or complex equipment needs to be added and Deep Learning. This article has also in characterization of pulmonary as the DL algorithms are applied to the considered the imaging procedure split lesions: approximation towards clinical images and in turn reduce the appearance into three elements, preparation, during indications. Ann Nucl Med. 2010 April of respiratory motion, delineating small and post procedure. When considering the 24 (3) 207-14 nodules otherwise unclear. It is important procedure broken down in this manner, 10. Artificial Intelligence and the however to make clear that the use of AI there are multiple points at which AI Radiographer/RT Profession: A joint does not diminish the requirement for the features to enhance the procedure and statement of the International Society Radiographer or RT to fully understand assist radiographers/RT’s, to ensure they of Radiographers and RT’s and the complexities of the imaging technologies can achieve the optimal image quality and European Federation of Radiographer that they operate. patient care they strive for. Societies, Radiography, May 2020 Volume 26, Issue 2, p93-184, e25-e128 AI in NM & MI - Post imaging procedure It is important however, to also recognise www.radiographyonline.com/article/ There are many reconstruction and post that AI technology has actually been with S1078-8174(20)30037-7/fulltext processing algorithms which can be us for many years. We need to continue to applied in NM and MI. It is important embrace technological advancements as
16 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 Elevating patient care with Artificial Intelligence: An educator’s point of view
By Yudthaphon Vichianin, Thailand
IN this year, the World Radiography Day in statistics, basic research methodology, theme is “Elevating Patient Care with and research ethics may be required in the Artificial Intelligence”. Radiographers are students’ classwork. For example, at the one of the key stakeholders in delivering Department of Radiological Technology, patient care with artificial intelligence in the Mahidol University, the third-year students fields of radiology, radiation therapy, and are required to work on the Term Paper nuclear medicine. Recently, a joint statement I and II (total two credits with in two of the International Society of Radiographers semesters) for literature reviews in various and Radiological Technologists and the radiology areas, research methodology European Federation of Radiographer and data collection techniques as well as Yudthaphon Vichianin Societies was published on this topic. The some basic statistics for data analysis in Dr Yudthaphon Vichianin is joint statement article can be directly different types of researches. Upon finishing the current ISRRT Director of accessed at the ISRRT website1 or on the third year, students are equipped with Education. He graduated from ScienceDirect2. The statement suggests a broad view on conducting research and Mahidol University with the degree that the “radiographers and radiological would be ready to perform a senior research of Bachelor of Sciences (B.Sc.) in technologists are the interface between project under their advisor guidance. The Radiological Technology in 1995 and imaging technology and patients.” It also AI techniques commonly employed and received his Master of Sciences in indicates the importance that radiographers published research publications can be Information Technology (MSIT) at and radiologic technologists have in the used as examples of research methodology Mahidol University in 2000. He was a process of AI algorithms testing and to introduce senior project classes to the Royal Thai Government Scholarship validating prior to clinical implementation. principles and application of AI technology. recipient to pursue his study in As the ISRRT Director of Education, I offer Master of Sciences in Information my personal perspective for educators to What new subjects will have to be System (MSIS) at the Hawai’i Pacific consider as they prepare their students to incorporated to have students University in 2003. He continued his deal with the AI era as they become educated begin to understand and how is AI doctoral degree in Communication AI users of the upcoming technology. technology being used in education and Information Sciences from the in radiography profession? University of Hawai’i, Manoa and What will educators around the world need Since the education program for graduated in 2007. to adapt their teaching for radiographer radiographers and radiological technologists students to prepare to AI in the future? varies from region to region, I would like to Currently, Yudthaphon is an The joint ISRRT EFRS statement states outline the topics for educators to consider assistant professor and the program that before using AI in clinical settings for incorporating into the program. In order chair of the Master of Sciences any proposal has to be validated before to broadly understand how AI works, some in Radiological Technology at the implementing it in practice. Radiographer backgrounds in mathematical and statistical Faculty of Medical Technology, and radiological technologist students modelling techniques related to AI are Mahidol University in Bangkok, should be prepared and equipped with an required beyond the basic courses. Hence, Thailand. essential skill set during their classwork mathematics and statistics classes are or in their senior projects the ability to considered important core courses in the Yudthaphon has focused his perform systematic approaches in basic program. The combination of these subjects professional expertise upon the research or tools/technology evaluations. is not only beneficial in AI but also important intersection of ICTs, radiation In the real situation, the evaluation of for scientific methodology in general. technology, and education. the AI prior to clinical practice may be preceded as research projects. The project Moreover, a large data set is usually He is highly active in the Thai Society teams may include physicians, engineers, required to test the developed mathematical of Radiological Technologists (TSRT) physicists, radiographers and radiological or statistical modelling in AI. As a result, giving numerous training classes technologists, statisticians, scientists etc. programming and data management in PACS and MIIA certification Working as a team of professionals, the skills are also essential skills. Python preparation. He teaches at a number role of radiographers and radiological programming language is one of the most of universities in Thailand as an technologists in the projects may vary from powerful computer languages and greatly invited lecturer. that of image acquisition, data collection, used in the AI development process. Also, or data analysis. As a result, some basics various tools associated with Python are
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 17 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW widely available for users and developers. ability to automatically learn and improve work assignments. For example, there is Hence, educators may consider including from experience without being explicitly a Massive Online Open Course (MOOC) in or adding some programming and data programmed in advance. Machine learning Data Mining using WEKA software offered management content into their courses focuses on the development of computer by the University of Waikato, New Zealand. in the curriculum. One great application programs that can access data and use It can be accessed via Youtube.com at of AI is Machine Learning (ML) technique. it learn for themselves. This is one of www.youtube.com/user/WekaMOOC. The According to the Expertsystem.com3, ML great resources that educators can use “WekaMOOC” offers a wide variety of video is an application of artificial intelligence in learning assignments and materials in as e-Learning that can be a great starting (AI) that equips any system with the their advanced course work or advanced point for students and individuals who are interested in AI. In addition, the “WEKA” open source software4 is also freely available and ready-to-use with an intuitive graphic user interface. It can be downloaded from www.cs.waikato.ac.nz/ml/weka/. If you want to learn more in depth techniques of machine learning using WEKA, a book titled “DATA MINING: Practical Machine Learning Tools and Techniques”5 is also available at additional cost.
For a more advance AI technique, deep learning6 has recently gained huge interest from scientists around the world. In deep learning, a more complexed modelling together with a larger data set is utilized to improve the system ability to deeply experience and make more accurate decisions. Luckily, at Kaggle.com7 there is a set of free access online courses for anyone who is interested in developing the data science skills. The courses include Python programming, introduction and intermediate level for Machine Learning, data management using SQL (Structure Query Language) both introduction and advance levels, and many more. These are invaluable resources that educators can access and utilize for their classes or use in any radiography and radiological technologist programs. These free online courses are well organized, and more importantly, they can be freely accessed anywhere and anytime upon your convenience.
One of the important roles of radiographers and radiological technologists incorporating AI in our professional practice is patient care. The joint statement of the International Society of Radiographers and Radiological Technologists and the European Federation of Radiographer Societies emphasize the important on the role of radiographers and radiological technologists as clinical decision makers. We should use AI as our support tool, “not a replacement for, clinical judgement and professionally accountable decision making”. This reinforces the fact that the AI is an extension to the professional capability.
18 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 In my opinion, patient care and 6. Deep learning [Internet]. En.wikipedia. www.kaggle.com/learn/overview communication skills still play a very org. 2020 [cited 26 July 2020]. Available 8. Bachelor of Science in Radiological important role in our professional practice. from: https://en.wikipedia.org/wiki/ Technology – Faculty of Medical AI may be utilized as tools such as image Deep_learning Technology [Internet]. Mt.mahidol.ac.th. enhancement, smart patient scheduling, 7. Learn Python, Data Viz, Pandas & More | 2020 [cited 26 July 2020]. Available from: or improved patient workflows. However, Tutorials | Kaggle [Internet]. Kaggle.com. https://mt.mahidol.c.th/en/academic- radiographers and radiological technologists 2020 [cited 26 July 2020]. Available from: programs-en/bsc-rt/ are “the interface between imaging technology and patients.” At my school, beside the technology related classes the students are also required to take “Patient Care and Radiological Service Management” classes8 which focused greatly on patient- centered services in clinical practice.
In summary, as an educator myself, I wish that all my colleagues around the world may consider and start empowering their students with essential skills including mathematics and statistics for AI, data management (i.e. SQL), computer programming (i.e. Python). I hope that these examples of available resources and information presented become your starting point to inspire your classes and curriculum development in preparing our next generation students to become great radiographers and radiological technologists in the near future whilst embracing the opportunities of AI in the modern era.
Reference 1. Artificial Intelligence [Internet]. ISRRT. org. 2020 [cited 26 July 2020]. Available from: www.isrrt.org/artificial-intelligence 2. The European Federation of Radiographer Societies. Artificial Intelligence and the Radiographer/ Radiological Technologist Profession: A joint statement of the International Society of Radiographers and Radiological Technologists and the European Federation of Radiographer Societies. Radiography. 2020;26(2):93-95. 3. What is Machine Learning? A definition - Expert System [Internet]. Expert System. 2020 [cited 26 July 2020]. https://expert system.com/machine-learning-definition 4. Weka 3 - Data Mining with Open Source Machine Learning Software in Java [Internet]. Cs.waikato.ac.nz. 2020 [cited 26 July 2020]. Available from: www. cs.waikato.ac.nz/ml/weka/index.html 5. Eibe Frank, Mark A. Hall, and Ian H. Witten (2016). The WEKA Workbench. Online Appendix for «Data Mining: Practical Machine Learning Tools and Techniques», Morgan Kaufmann, Fourth Edition, 2016.
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 19 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW
Elevating patient care with Artificial Intelligence
By Denise Choong, Singapore
ARTIFICIAL intelligence (AI), as the name lesions in the training dataset. This can suggests, could be broadly described as be particularly useful with indeterminate the programming of a machine to emulate lesions or when there is wide intra- and human-like functioning. Many would inter-observer variability in the scoring to associate this with robotics and machine reduce unnecessary false-positive biopsies automation and fear that the increasing that can increase health care costs and adoption of such technology could patient anxiety. ultimately replace us. The task of regression allows for AI in Ultrasound continuous output values as opposed to Denise Choong There are four main types of applications in categorical classification and has been Denise is a senior radiographer at AI in ultrasound: classification, detection, used in obstetric scans. Often time due National University Hospital (NUH) regression and segmentation. to difficult positioning of the foetus, from the department of The most notable applications would it is challenging to obtain dimensions diagnostic imaging. She graduated be classification with computer-aided in the standard plane. By using AI to with a Masters of Science in diagnosis (CAD) and the detection of determine the standard plane of the Ultrasound from University lesions. Majority of the developments foetus to facilitate measurement, the College Dublin (Ireland) and has have been made in breast, thyroid and abdominal circumference or 3D ultrasound been working at NUH for 9 years. liver lesion detection based on widely information of the foetal brain can She performs general, used radiological classification systems be effective in estimating gestational paediatric and vascular ultrasound such as the BI-RADS scoring for breast age, even when the foetus is not in the scans and has a special interest in lesions (Akkus et al., 2019; Brattain et al., optimal scanning position. This can avoid musculoskeletal (MSK) 2018). By recognising echotextural and extended scanning times or inaccurate ultrasound. She currently morphological parameters like shape, measurements that could lead to possible facilitates the advanced training of margins and acoustic shadowing, AI misdiagnosis (Liu et al, 2019; Brattain et juniors in MSK ultrasound. classification applications can differentiate al., 2018). Denise is in charge of the EOS low types of tissues. Detection functions can dose X-ray system at NUH used identify and locate objects of interest Segmentation is primarily used to identify for dedicated long fil scoliosis such as lesions. Together, they can structural boundaries of anatomical and lower limb imaging. She also provide radiologists with a second opinion and pathological tissue. It allows auto retains a keen interest in intra- through suggesting a score based on delineation of structures which is operative radiography. known features of other similar appearing particularly useful in non-rigid organs such
Denise is the President of the Table 1: Brief summary of tasks in AI for ultrasound Singapore Society of Radiographers and the Singapore representative AI task Function/Uses Application of the International Society for Classification Differentiate types of tissues into Classification of breast lesions Radiographers and Radiological discrete categories according to BI-RADS Technologists (ISRRT). classification; Thyroid nodules
Detection Identifying and localising objects of Locating breast, thyroid and liver interest such as tumours and lesions lesions
Regression Estimates continuous values Estimations of gestational age according to abdominal circumference or 3D images of the foetal brain
Segmentation Demarcation of structural Outlining the ventricles of the boundaries heart; developmental anatomy of the infant hip; thyroid nodules
(Akkus et al., 2019; Liu et al, 2019; Brattain et al., 2018; Quader et al., 2017)
20 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 as the heart (Liu et al, 2019). Segmentation good diagnostic confidence regardless system. Furthermore, at the expense of can also outline the anatomical structures of sonographer experience. Improved efficient machine learning predictions, of an infant hip to facilitate measurements sensitivity of AI in the future may also there may be reader bias of the predicted using 3D ultrasound acquisitions (Quader confer greater reliability in minimizing findings which can result in reinforcement et al., 2017). This presents a solution for missed diagnoses at the point of care. of findings that may be false. less experienced sonographers who may For experienced sonographers, AI could find it challenging to acquire high-quality improve imaging workflow by suggesting Another common concern in AI-based images with fretful infants. and modifying the user interface to make CAD is that it is unable to explain how it relevant tools more accessible, thereby derives at its decision. The processing These tasks may seem similar and are providing diagnostic and decision support. ‘black box’ introduces ethical complications not always mutually exclusive; they are The system could also employ auto- with regard to whether the manufacturer often prerequisites for the other. The annotation to reduce labelling errors that or radiologist should be held responsible brief examples of AI applications in if gone unnoticed, may drastically impact in the event of a misdiagnosis. Such ultrasound mentioned here are by no patient care downstream (Kusta, 2018). ethical and legal conundrums need to be means exhaustive. Research into its uses In addition, prediction by AI can prioritise resolved to implement AI safely with proper continues to expand rapidly to further worklists by identifying high-risk patients governance (Ho et al. 2019). improve the diagnostic capabilities of (Matthew, 2019) and prompt further ultrasound machines which ideally would investigation to alert radiographers. As Singapore has taken a cautious stance, translate to better patient care through such, patients can receive early intervention starting with emphasis on supporting accurate diagnosis with shorter scanning to mitigate the risks of deterioration, the infrastructure for AI. In addition to times and fewer invasive procedures. ensuring that high patient care standards developing and augmenting expertise are preserved. locally, a model AI governance framework AI in the patient journey was established to safeguard against the While many of these developments have The fact that ultrasound is an operator misuse of AI. It includes guidelines on how focused on aiding radiologists in diagnosis, dependent modality makes the human to address such ethical and governance AI-powered capabilities can impact patient touch an intrinsic part of any sonographic issues including determining the level care at various parts of their patient examination. AI can supplement our of human involvement, ensuring proper journey in the imaging department. pathology detection capabilities and internal governance through appropriate clinical decision making skills but it standard operating procedures and training Another subset of AI that is often cannot replace the communication and of staff. The need for transparency and overlooked is planning and automation. rapport we establish with our patients fairness in AI decision making is also With automation in scheduling patients, during the examination. Furthermore, the highlighted (IMDA, 2020). appointment-based services may see required skin exposure of the body for the smoother patient traffic. AI-powered scan makes our attention to the needs Conclusion analysis of current data can also optimise of the patient paramount in providing a Although there are many issues that scheduling and workflow to streamline professional and comfortable setting for still need to be resolved regarding the administrative processes. This can help to the scan. implementation of AI in the healthcare minimise patient wait times and provide system, the abundant research to date them with a more seamless and pleasant Challenges of AI adoption has shown great potential, particularly in experience. Furthermore, inefficiencies in Despite the exciting promise of various radiology. The systems are complementary scheduling and manpower planning can AI capabilities, premature adoption of the to our radiographic practice to enhance be a precursor to accumulated workload. technology when the infrastructure and diagnostic capabilities, but it definitely Consequently, radiographers often macro-environment are not yet established cannot replace the radiographer’s human experience large patient loads with skeletal is risky. As the success of AI is highly touch and the need for patient care. manpower leading to fatigue and increased dependent on the accessibility of data, With the advancement of technology, stress. This could be mitigated with the having proper governance and data control radiographers need to be prepared to use of AI in planning, to improve the patient regulations is vital for the appropriate receive the advent of AI. With greater experience and look out for radiographer’s handling and sharing of such highly awareness of research in AI, its potential wellbeing to ensure they are physically sensitive healthcare data. pitfalls and perhaps a basic understanding and mentally healthy to care for their of how AI works, we will be better equipped patients well. The data itself needs to be processed to utilise the systems as their prototypes appropriately before it can be used. are rolled out. AI image recognition capabilities can With a large mix of structured (largely potentially accelerate the training of quantitative) and unstructured (medical Finally, it is important to remember that new sonographers to identify anatomical images, free text clinical reports) data patient care not only revolves around the structures and pathologies. These systems scattered over many locations, the act of care itself but also encompasses can be programmed to help improve challenge for AI developers is to reliably being responsible, knowledgeable and detection of potential lesions in real-time and accurately extract and produce ‘clean’ to ensure that scans are performed with data in a recognisable format for the continued on page 23
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 21 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW Artificial Intelligence (AI) in radiation oncology, from perspective of a radiation therapy manager
By Jo Page, Australia
MY first thoughts when preparing to information based on protocols with certain write this article were of course “What is objectives and constraints. The current Artificial Intelligence? So, as we all do, I iteration in a TPS is where we have seen went to a search engine and asked that the increasing use of AI. Knowledge Based question, and of the 837,000,000 answers I Planning (KBP) uses a database of previous went with Wikipedia, for no reason except treatment plans to create a predictive that it seemed to be the simplest yet most model for different treatment sites. To use relevant description pertaining to the KBP each department uses their previous subject of this article. treatment plans to create their own model, which can be extremely time consuming Jo Page Artificial intelligence, sometimes and sometimes subjective passage of work. Jo is currently the Director of called machine intelligence, is Departments must ensure that previous Radiation Therapy in the Radiation intelligence demonstrated by plans adhere to their current evidence- Oncology Department at the Chris machines, unlike the natural based protocols, which will produce a O’Brien Lifehouse (COBLH), in intelligence displayed by humans model that is applicable to their existing Sydney Australia. and animals. Wikipedia practice.
Jo’s role includes overseeing How does this relate to radiation therapy, As these iterations of treatment planning all aspects of running a busy how does it improve what we do, does have evolved there has been an emphasis Radiation Oncology Department, it make us a more efficient profession, on improved treatment quality, safety the management of a staff of 46 but most importantly does it distract or and efficiency within departments. It is radiation therapists, and is part takeaway from the most important area of imperative for patients to begin their of the management team, which our profession which is patient care? course of treatment with as little delay as includes all the core professions, of possible, and KBP planning is proving to be the department. Firstly, I will outline some of the ways that an efficient method of achieving this. AI tools are used in radiation therapy. I feel COBLH is a not-for-profit integrated it is important to note that these are my For a treatment plan to be generated it cancer treatment centre and interpretations of their uses, not all will requires tumour volumes and structures hospital, and Jo is also part of the agree, and of course there are other areas to be contoured from a CT image dataset. Senior Leadership Team for the that some may consider are sources of AI in As treatment techniques have evolved organisation. our profession. requiring higher doses to the tumour, complete and accurate tumour volume Jo has been a Board Member and As we know, the aim of any patient’s and structure delineation is required. This Past President of the Australian course of radiation therapy is to deliver the process is time-consuming and can be Society of Medical Imaging and necessary dose to the tumour, with as low affected by inter-observer variation. To Radiation Therapy (ASMIRT) as possible dose to surrounding tissues and improve this contouring process, Atlas and is currently the ASMIRT organs at risk (OAR), with the end-goal of Based Auto-segmentation (ABAS) software Representative on the ISRRT maximising tumour kill while at the same has been developed and incorporated Council. time reducing toxicity to the patient. into Treatment Planning Systems to Treatment Planning and the use of automatically contour OAR volumes. The Treatment Planning Systems (TPS) in process uses the knowledge of prior image radiation therapy has evolved from 2D sets with labelled structures. The basis hand-drawn planning, to conformal 3D of this requires the generation of an atlas forward planning (with the introduction of of patient contours into a library with use of CT scanning), to 4D inverse planning the relevant structures then aligned and commonly using Intensity Modulated deformed to the anatomical structures of Radiation Therapy (IMRT) and Volumetric the treatment patient. Arc Therapy (VMAT). Inverse planning uses
22 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 One of the more recent advances in implementation of KBP has improved in our hands. radiation therapy using AI, is Adaptive efficiencies in the planning process, Radiation Therapy (ART). “ART is a radiation which transpires to patients starting their The Australian Society of Medical Imaging therapy process where treatment is treatment earlier, and I am sure this is the and Radiation Therapy has introduced an adapted to account for internal anatomical case in most departments that have KBP. AI professional working group, that will changes. Some organs in the body can shortly produce guidelines around the change in size and shape over the days and It is important to remember that all use of Artificial Intelligent tools in the weeks during a course of treatment. The systems such as KBP and ABAS have a workplace. aim of ART is to account for these changes human element involved in their design and and deliver the radiation dose to the implementation for individual departments. References tumour as accurately as possible.” (Peter KBP models are designed using plans Implementing user-defined atlas-based MacCallum Cancer Centre, n.d.). previously based on departmental auto-segmentation for a large multi- protocols. The same applies to the centre organisation: the Australian The process of ART is to view the day of development of an Atlas, which relies on Experience. Yunfei Hu BMedPhys, treatment imaging, access the patients contouring done manually, whilst by well MSc(MedRadPhys) Mikel Byrne BSc, diagnostic imaging, create AI contours, qualified professionals, but who none the MMedPhys Ben Archibald-Heeren calculate the dose for the scheduled plan less may have different interpretations of BMedRadSc(RT), Msc(MedRadPhys) et al on that day’s anatomy, create an AI driven the same image. Clinical Evaluation of Commercial Atlas- adapted treatment plan, evaluate, check Based Auto-Segmentation in the Head plan and deliver. There is no doubt about the rigidity and and Neck Region. Hyothaek Lee,† strict quality controls that are involved in Eungman Lee,† Nalee Kim, Joo ho Kim, The question being asked is does the use any new product development, however Kwangwoo Park, Ho Lee, Jaehee Chun, of these AI tools elevate patient care in any tools that we use are only as good Jae-ik Shin, Jee Suk Chang,* and Jin the radiation oncology arena? The answer the decision-making processes of those Sung Kim* is positive of course, as anything that involved, which can be highly subjective and Ethos therapy redefines personalized increases patient survival and/or improves involve, to some extent, a value judgement. cancer care. An Adaptive Intelligence™ their time being cancer free, without long- As medical radiation practitioners working solution, term side effects and good quality of life, in the highly emotive area of radiation www.varian.com/en-au/products/ is the final goal. As systems such as ART oncology we must never forget the adaptive-therapy/ethos: are improved and become more efficient, importance of the patient and their needs Peter MacCallum Cancer Centre. (n.d.). they will become mainstream. However, it both physically and mentally. Technically Adaptive Radiation Therapy. Retrieved is often difficult to implement such systems we must offer the best care possible, but August 3, 2020, www.petermac.org/ in a busy radiation therapy department if we must never overlook the importance services/treatment/radiation-therapy/ it has a detrimental effect on the patient of the individuals that are under our care, other-advanced-technologies/adaptive- workflow. In my own department, the whom, to some point, are putting their lives radiation-therapy
continued from page 21. College of Radiology, 16(9), pp.1318- medical-imaging/artificial-intelligence- 1328. within-ultrasound/ informed radiographers to remain Artificial Intelligence (2020) Infocomm Liu, S., Wang, Y., Yang, X., Lei, B., Liu, L., indispensable in elevating patient care. Media Development Authority, accessed Li, S.X., Ni, D. and Wang, T., 2019. Deep 5 July 2020, www.imda.gov.sg/ learning in medical ultrasound analysis: Acknowledgements infocomm-media-landscape/SGDigital/ a review. Engineering, 5(2), pp.261-275. I would like to thank Dr Jeevesh Kapur, tech-pillars/Artificial-Intelligence Matthew J (2019) The Society of Dr Andrew Makmur, Harris Abdul Razak Brattain, L.J., Telfer, B.A., Dhyani, Radiographers, accessed: 3 July 2020 and Ang Xu Kai for their contributions to M., Grajo, J.R. and Samir, A.E.,
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 23 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW Artificial Intelligence and imaging informatics
By Alexander Peck, United Kingdom
WITHIN the Medical Diagnostic Directorate, in control centres to determine the extent Imaging Informatics (the use of computers and location of leaking water or gas pipes in and to support patient care) is most likely - the system simply ‘listens’ to normal the place where many individuals believe AI background noises and activities, ‘learning’ is already highly integrated and prevalent what is normal and what may indicate in daily work. While it is true that many an abnormality. A leak or break can then countries are investing millions of dollars be flagged up in real-time to engineers. into AI research the view from the offices This leads to the question: Why is AI not of PACS Managers and other professionals more widely utilised within healthcare in the speciality around the world is at the presently? For example, the modelling of Alexander Peck moment more inconclusive on the true blood flow is remarkably similar to some Alexander Peck is a Superintendent distribution currently available. fluids frequently monitored by the utility Radiographer and Chartered IT companies, particularly when examining Professional based in London, UK. Presently, in the field of Informatics, the turbulent nature of flow within the common implementations of true AI are heart versus the turbulent flow of water in He is the current Chair of the fairly difficult to come by. Many currently a pump value. 4D, motion based imaging Society & College of Radiographers used systems touted as containing AI (for is already available in research settings IM+T Advisory Group, example some breast nodule detection and many dedicated Radiographers have Communications Lead for the and alerting systems, orthopaedic bone examined ways in which more subtle British Institute of Radiology and recognition and labelling algorithms or appearances on Cardiac MR and Cardiac past Vice President for Informatics common chest pathology classification Fusion studies can be highlighted for closer of the UK Radiological Congress. add-ons to PACS) actually be better analysis - preventing minor details form labelled Computer Aided Detection being overlooked by Reporting personnel. He is renowned for his extensive systems, and don’t truly have the capacity These types of imaging being available international voluntary educational to ‘learn’ and adapt as they ingest more provide the perfect raw data for an AI work in the imaging informatics studies. There is nothing wrong with this, technology to work with, likewise additional field, from devising and running indeed it is a similar story with ‘self-driving sensors could be developed to form part the UK’s only national non-profit cars’ such as Tesla Motor’s ‘autopilot’ of existing therapies such as implantable informatics training for PACS feature across its range still requiring a cardiac devices to provide a data source Teams, Radiology System Managers significant degree of human control with for detecting cardiac problems ‘at source’ and Allied Health Professionals minimal on-board learning and adaption and in real-time rather than requiring an over the past decade to training features. This effect of potentially over out-patient or hospital intervention. The thousands of radiographers on the describing medical software features approach to normalising AI in healthcare fundamentals of the speciality. as ‘AI’ (for either marketing or ‘hype’) seems to need an adjunct though and this at present has the potential to have the may be the patient themselves: could AI Alexander lectures widely in effect of reducing the perceived value of be utilised along with the data provided by universities and at medical the true AI systems, which as many others patient worn sensors to perhaps provide conferences and is also the author have described have the power to greatly those same Cardiac Radiographers more of a number of texts in the global affect patient care and simply the workload contextual data from their patients? ‘Clark’s’ radiology series of books, burden on Radiographers in the (potentially Fluctuations of blood pressure, blood including Clark’s PACS, RIS & near) future. biomarker status and user input activity Imaging Informatics (pacsbook. (exercise etc.) classifications would com). Numerous positive examples of the AI complement the data collected by sensors technologies finding well lauded and highly to allow for more detailed analysis and a beneficial usage have been demonstrated ‘wider picture’ view. outside of healthcare - for example utility supply companies (such as water, fuel or This however leads to the perennial gas suppliers) are able to use Artificial question of accountability and liability. Intelligence networks to process signals Returning to ‘self-driving’ motor vehicles, from IOT (Internet of Things) sensors many justifications around the world have embedded under pavements, roads and yet to rule definitively (due to lack of test
24 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 cases) on who would be responsible for a professionally developed service, making evolution of the technology - an evolution a fatal car accident involving a vehicle this is less common process than it could which will no doubt be required to under AI control – the ‘supervisor’/ otherwise be. Allowing AI applications overcome the legal hurdles posed in our driver, the software developer or the car to take care of some elements of this various jurisdictions around the world. manufacturer? These same questions will journey (notably if it were able to ‘clear’ Being very much in its infancy, Artificial be applied to any new AI implementation non-afflicted patients reliably and allow Intelligence has the potential to change within healthcare – an historically far more for clinical time to be better preserved for the way radiology works, but likewise it litigiously contentious environment. The those with injuries) would be innovative has the potential to be a large consumer use of AI has inherent legal implications and allow for radiographers to focus more of financial resources if radiographers given the underlying loss of human control, on the human aspects of the interactions working on the frontline are left out of the but also in the provision of training data: rather than the time consuming reviews, decision making and scoping processes. all AI algorithms must first work on a paper trails and referrals which cause the large training dataset in order to become phenomenon known as ‘imaging drag’ (the ‘intelligent’. Where does this dataset come delays caused to a standard patient journey from? Patients may not consent to the use due to administrative processes or non- of existing medical data for this kind of optimised uses of technology) today. ‘machine training’ purposes. In addition, there are also ethical questions: could AI be Other areas of application of AI which ‘trusted’ to automatically prioritise cases would benefit the radiology department on a busy reporting worklist depending on are more mundane, and most likely what the AI believes may be demonstrated to be thought of as less exciting than (or not) as the pathology even if this were some of the other articles in this special to delay other patient care or change a edition! One such example is the use of treatment pathway? On the other hand, AI in examination justification (vetting) as reporting backlogs continue to rise or protocolling (assigning a study to the around the world, the implementation correct speciality, area or modality). of AI in worklist control could even take Countless hours of radiology department over the entire function of reporting less time are taken by these few crucial steps significant caseloads and allow reporting found in every imaging examination request Radiographers to become more focussed and yet very little has historically been on developing speciality reporting services offered to automate or streamline these. (CTs etc.), presenting an opportunity to Allowing AI to take over the vetting and increase patient care if implemented and protocolling roles would allow for staff controlled well. time to be more focussed on imaging and the actual physical patient-care elements, Another such innovation which could be but with the topic being so unremarkable most likely benefit from the development of it is unlikely to attract funding or AI assistive technologies are Radiographer researchers into progressing it – a key Led Discharges. This process, which prerequisite to development. occurs in limited extents around the world, is where a radiographer is responsible Of course we have to ask ourselves, as for the majority of the patients entire with many IT based technologies, is pathway through the healthcare service AI a solution looking for a problem? and is likely to be successful in bony Presently, all AI based or machine trauma situations where referrals are learning algorithms must only be used comparatively simple (patient history is as a ‘diagnostic aid’ - final interpretation available, the clinical context is known, being performed by a licensed healthcare the singular clinical questions being practitioner with the appropriate training. generally: is there a fracture? etc.), imaging Are we simply looking for applications of typically ‘while-you-wait’ and reporting, this fascinating technology that really will evaluation or commenting protocols are still need a human? in place anyway to support immediate medical care to any detected abnormality. Overall, when considering these few Adding the further final step of dispatching examples, AI is at present underused the patient home (with follow-up clinic and primarily in place for computer aided appointments if for instance a fracture was diagnosis purposes. Radiographers detected and treated) is not revolutionary, promoting AI to completely handle but requires highly skilled and specially automated routine and time-consuming trained and authorised radiographers and tasks will be the ‘next big step’ in the
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 25 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW ALARA AI
By Noeska Smit, Netherlands
WHEN I was still actively working as a for data exploration, analysis, or radiographer in the Netherlands, about 15 communication purposes. years ago, I experienced the beginning of the transition from developing X-rays on The advent of artificial intelligence (AI) film to fully digital systems. This transition may lead to another turning point in the brought drastic changes to the workflow. workflow for radiographers. As previously Where I used to go to the radiologist to with the transition from x-ray films to discuss images film-in-hand as needed, digital solutions, AI-based software this changed to calling the radiologist and algorithms may play a larger role in the asking him or her to bring up the images. future in handling the technical aspects Noeska Smit We went from charming hand-written, of the job, for example improving image Noeska Smit has been an Associate at times in illegible doctors handwriting, quality or automatically indicating Professor in Medical Visualization paper X-ray requests to a fully digital potential pathology. In AI, typically at the University of Bergen, Norway, system. In addition, the devices became software is trained to perform certain since 2017. She is also affiliated more sophisticated. Where before I tasks based on a learning procedure. For to the Mohn Medical Imaging and was always thinking about picking the example, the computer takes hundreds of Visualization (MMIV) centre as a optimal film size, kilovoltage (kV), and labeled CT-scans and can then attempt senior researcher. milliampere per second (mAs), now it to automatically label an unseen CT- became possible to pick X-ray foot’ from a scan, based on what it has seen in the Currently, she is researching novel protocol database and maybe make small other labeled scans. While these days interactive visualization approaches adjustments from there. Of course, artificial intelligence is sometimes seen for improved exploration, analysis, this could reduce the amount of human as some sort of magical solution that can and communication of multimodal errors, such as incorrectly entering the solve almost everything, I would argue medical imaging data. wrong kV and mAs, but it also may it cannot and for the moment should be increase error if a radiographer takes critically examined before accepting it into Noeska is a licensed radiographer incorrectly recommended values without radiographer workflows. There are serious with three years of practical further thought. issues with many artificial intelligence- experience in the Netherlands at based solutions and I believe radiographers the MCRZ hospital in Rotterdam, This was also the time I quit my job as can play a key role in responsible use of AI. the Netherlands. a radiographer. I realized I was more interested in 3D CT reconstructions than in The first key issue with AI-based solutions She completed her studies helping orthopedic patients on the table. is that they operate as a so called ‘black in Computer Science at the I really wanted to dive deeper into these box’. This means that it is not possible to Delft University of Technology, more technology-focused aspects. see how the software came to its decision. specializing in Computer Graphics To follow up on the previous labeled CT and Visualization in 2012. In 2016, Another challenge that made my life as a example, AI will be able to provide the she obtained her PhD in medical radiographer difficult is that I had a hard labels, and potentially even how sure it is visualization at the same institute time not continuing to think about the more of these labels, but it cannot tell you how it in collaboration with the Anatomy severe cases I saw on the job in my private did it. This may lead to strange situations, and Embryology department at life. I went on to study computer science for example, closer examination of an the LUMC in Leiden. Her PhD and specialized in computer graphics and algorithm trained to identify pictures of research project, entitled ‘The visualization. I guess this sounds like a airplanes revealed that it actually identified Virtual Surgical Pelvis’ resulted in direction as far away from patient care these pictures only by looking for blue sky a detailed virtual 3D model of the as possible, but while my daily work life in a photo. For medical imaging data, we pelvic area from heterogeneous has changed a lot, my main goal is still should make sure that AI-based solutions anatomical data that is used in the same: I want to work on things that are trustworthy. It is very hard to trust in education and has potential for pre- can help people. In medical visualization techniques when it is completely unclear operative surgical planning. research, I work on developing systems how they come to their decisions. that make visual representations of medical imaging data that help people Another key issue is that AI-based carry out their tasks more effectively algorithms can make mistakes. Incorrect
26 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 results may be alright when it comes to sure the patient has the best possible identifying pictures of cats or gender- experience and undivided attention. Still swapping selfies but are unacceptable I don’t think we are quite there yet with when it comes to algorithms proposing the current state of AI and for the moment diagnostic or treatment advice. A am reminded of the guiding principle for radiographer has the skillset to check the radioprotection: As Low As Reasonably original image data against AI-enhanced Achievable (ALARA). I propose we keep image data to ensure that no mistakes the use of AI in radiographer workflow are made. Coming back to the CT labeling as low as reasonably achievable as long example, incorrectly placed labels could be as key critical issues are not addressed. spotted and corrected by a radiographer. The profession of radiography has a For AI-enhanced increased resolution, a rich combined expertise in technology radiographer could check that important and patient care. As such, I believe pathology is not de-emphasized. Here, the radiographers are excellently positioned technical expertise and experience that to be critical of newly introduced AI-based a radiographer has is valuable to verify solutions and responsible for ensuring that correctness. the best possible patient care is provided.
In my previous medical visualization research, we developed a technique that can visualize autonomic nerve regions in MRI scans for pre-operative planning, while these nerve regions are not visible in MRI originally. The goal here was to reduce severe surgical side-effects that arise from nerve damage. A surgeon could take this information into account in his or her planning, but the goal was never to let the computer do the planning or even the surgery. In this case, people are using computers to support decision making. We combine what people are best at with what computers are good at. Computer-assisted decision support is by no means new, and airplane pilots have relied on computer- guided controls for years. Key to the success of such approaches is to combine human expertise with computer support. In a way, I think radiographers could have a similar role with AI. We could combine what radiographers are best at with AI-based solutions for the more tedious parts of the job. These AI-based solutions can only be brought in if patient safety and treatment quality can be guaranteed.
The previously mentioned limitations of AI are under active research for the moment, trying to improve algorithm performance, as well as opening the black box by so-called explainable AI. If artificial intelligence becomes good enough at some point to take over some of the technical aspects of a radiographers job, this is also an opportunity for improved focus on patient care. If less time needs to be spent tweaking imaging parameters and acquisition protocols, this may lead to more time that can be spent making
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 27 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW
Breast Imaging and Artificial Intelligence
By Jill Schultz, USA
I SUSPECT I am not alone when saying I It adds marks to images noting any have had mixed feelings about Artificial finding that may seem suspicious. CAD Intelligence and mammography over the uses pattern recognition algorithms years. The term ‘artificial intelligence’ to determine if calcifications or mass- can seem somewhat unconventional like densities have suspicious imaging or futuristic. I often think of ‘artificial characteristics. It is up to the radiologist to intelligence’ as being referred to in various decide to agree or ignore these markings. action-filled science fiction movies. Only AI based CAD can help with obscure a couple years ago, I recall hearing findings that may have otherwise been radiologists talk about Artificial Intelligence missed and it can be a helpful decision Jill Schultz (AI) in a general manner as they wondered tool if there is a particular finding that the Jill holds a Bachelor’s degree where the future of AI would take them. radiologist is unsure of. in Business Management from Their discussion included wondering if AI the University of Sioux Falls, SD would replace them in the future, while As CAD helps determine what is wrong with and is a radiologic technology surpassing their human abilities. an image, it has been met with its share program graduate. ARRT It is that type of fear-provoking of criticism. Comments have ranged from professional certifications include conversations that makes me think the too many distracting marks on the images, radiology, mammography, quality term ‘artificial intelligence’ is somewhat to radiologist disagreements with many management and is a certified unfortunate. Perhaps it would be met markings, to a lack of reimbursement. navigator in breast management with less fear or resistance if this type of Over the years, some questioned how through the NCoBC. technology was referred to as ‘intelligent much CAD truly helped them. For some, decision assistance’ or something that traditional CAD became a bit of a nuisance Jill is the current Breast Program sounds more helpful. After all, the goal of producing a higher rate of false positive Quality and Accreditation AI is to improve patient care by increasing marks. False positive markings increase Coordinator at Avera in Sioux accuracy and providing efficiencies. It the risk of unnecessary diagnostic call Falls. Her professional experience is technology that can assist the overall backs and workups leading to possible includes managing/directing imaging process and it directly helps biopsies. Historical AI CAD technology is radiology, mammography, radiologists, which in turn ultimately not necessarily a time saver knowing the radiation therapy, breast surgeon benefits the patients we serve. radiologist needs to study the markings to clinic and various aspects of discern the risk of each. Regardless, many development of breast programs. Forget the fear- reality reveals there are facilities continue to use a form of artificial She has led multiple types of very practical roles for AI in breast imaging. intelligence via traditional CAD. breast and oncology program Whether you realize it or not, artificial accreditation efforts over the years intelligence has been an important part of Computer aided detection adapted as as well as project and program breast imaging for a number of years. mammography technology progressed from management, safety and service analog, to digital (2D), and now to digital excellence. She has served on Mammography breast tomosynthesis (DBT, 3D). Traditional professional local, state and With mammography, AI based computer CAD works in conjunction with analog national committees and boards aided detection (CAD) has been an and 2D mammography. The principles of with SROA, NCoBC, ASRT, ARRT, integral part of breast imaging for 20 AI CAD have changed with DBT and the SDSRT, grant committees, and years. Computer aided detection, CAD multitude of images that 3D mammography volunteer Haiti mission work since research started in the mid 80’s and the produces. The uses and benefits of artificial 2012, helping to start a breast first mammography CAD was available intelligence with DBT mammography has program in Western Haiti. She has for purchase in 1998. CAD technology has flourished. provided numerous breast health continued to change and greatly improve public and professional educational its performance over the years, always With the production of additional images presentations and training for new maintaining the goal to improve accuracy per exam, DBT interpretation time mammographers and authored as an aid to radiologists. This decision increased to nearly double. The focus of several breast health articles/ support tool acts like a ‘spell checker’ that DBT artificial intelligence goals has truly documents. the radiologist can use after the initial case expanded to reduce false positive and review but prior to their final interpretation. false negative rates as well as provide
28 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 efficiencies while increasing overall which patients may benefit from addition detection rate. This is exciting for the diagnostic work ups. By enhancing future of breast imaging. relevant regions of interest it also saves interpretation time. Reducing overall Additional Breast imaging interpretation time, whether seconds or Breast imaging artificial intelligence is minutes per exam, provides more time for used well beyond mammography. Typically, patients and increased imaging throughput. quality breast magnetic resonance imaging Efficiencies enhance report turn-around (MRI) includes the use of CAD. MRI CAD –time while increasing patient, radiologist, helps to evaluate suspicious breast lesions technologist satisfaction. AI tools like for the radiologist. At this time it is unusual this have the potential to help prevent not to use CAD with breast MRI. radiologist burnout. AIso, with the wide reports of breast imaging backlogs due to Progress to reduce interpretation time the pandemic, AI time saving advances is and add efficiencies continue to expand. welcomed technology. AI tools are being developed with breast ultrasound. Reducing interpretation time Breast imaging and AI’s future helps to allow for more time with patients Artificial Intelligence technologies are and increase throughput of cases. As changing rapidly as the benefits and uses contrast enhanced spectral mammography, of AI are being realized. I look for further CESM technology continues to become expansion and automation of the use of AI more widespread, the opportunity to in breast imaging over the next few years. include more AI features emerge. The future will require us to have an open mind and to think ‘bigger picture’ on how AI Impact of Artificial Intelligence can help. Artificial intelligence is rapidly developing while exploring how it may ultimately Price points will be key as various breast improve patient outcomes. MRI, CT, imaging artificial intelligence products Interventional and diagnostic radiology are become available. The economic benefits just a few additional imaging modalities of AI needs to be considered. Cost must be that AI benefits. With the current pandemic, weighed against value, overall savings and there are additional opportunities for AI reimbursement. technology to aid with COVID 19 detection and monitoring. Just as education is important to professionals it is also helpful for the Specific to breast imaging there is promise patients. If AI is used for their imaging for AI to provide a reduction of image noise, exam, explaining the role it played in the artifacts, motion and dose. Increased diagnosis is important. Explaining that AI resolution improves overall imaging quality. works to assist the skilled radiologists is The synthetic, or computerized images being honest and informational. used with DBT makes such image quality improvements a reality. I feel the most potential for AI technology is with it used as a second, or pre-read, Mammography AI options now include and prioritizing cases. Focus on triaging positioning feedback and review and cases that are in most need of reading first scoring of breast density which is will be very beneficial to patient care and associated with risk for breast cancer. This those most at risk of breast cancer. Today, density feature removes subjectivity from now more than ever, healthcare is faced this score while reducing variability. It can with workflow challenges and increased help the radiologist determine the most volumes making this feature valuable. appropriate density level when challenged with a case that seems borderline, between Whether Artificial Intelligence is fear two levels of density. provoking or openly embraced, it is here to stay and will only continue to assist the With the onset of large data set, deep multi-modalities of breast imaging and learning technology, which becomes improve how we care for our patients. smarter as it reviews more images, AI can help predict the level of suspicion of a breast nodule. This helps determine
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 29 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW
Artificial Intelligence in cardiology
By Christopher Steelman, USA
AI Is revolutionizing the under the guidance of a fluoroscope, practice of medicine he advanced the catheter 65cm into his ARTIFICIAL intelligence (AI) is into the right atrium.4 Dr. Forssmann revolutionizing the way we live, work, had the radiographer capture an image and communicate. It autocompletes the obtaining documentation of his incredible sentences as we type, populates internet experiment. The radiographer and searches before we complete our thoughts, cardiology have been inseparable ever allows cars to drive themselves, filters since. Today, more than 2 million patients unwanted email, and supports language are treated with coronary angioplasty Christopher Steelman translation. AI is also revolutionizing the annually.5 From the beginning, the Christopher is passionate about the practice of medicine. It’s assisting doctors radiographer has played a fundamental role of technology in education. He is to diagnose patients more accurately, role in the multi-professional teams who a regional coordinator of professional make predictions about patients’ future treat patients with life and limb-threatening practice for the ISRRT, the chair of health, and recommend more effective diseases. Contributing far more than an the World Radiography Educational treatments. There is an increasing number expertise in imaging, today’s cath lab Trust Foundation, and chair of the of exciting applications in the medical radiographer has accepted the challenge ASRT Foundations International imaging field, propelling it forward at a of rapidly developing technology and AI will Outreach Review Committee. rapid pace. AI can help clinicians diagnose not be the exception. disease and optimize treatment processes, He has addressed audiences at reduce the rate of misdiagnosis, and Global cardiovascular disease the ISRRT 19th and 20th World improve diagnostic efficiency. AI can now Cardiovascular disease (CVD) is the leading Congress in Seoul, Korea, and recognize medical images and provide cause of mortality worldwide, accounting Trinidad & Tobago, the ASRT clinicians with more reliable imaging for >17.3 million deaths per year in 2013,6 Educational Symposium, the diagnostic information. And AI is also a number that is expected to grow to >23.6 RAD-AID Conference on Radiology accelerating research by giving us the million by 2030.7 in Developing Countries, and the ability to analyze extremely large data sets Association of Educators in Imaging that were previously difficult to analyze Deaths attributable to ischemic heart and Radiologic Sciences. He has using traditional data processing methods. disease increased by an estimated 41.7% contributed to numerous publications Artificial intelligence and machine from 1990 to 2013.8 Coronary heart including Radiology in Global Health learning (ML) are poised to influence disease (CHD) is the single largest cause and The Heart Revealed, a project of nearly every aspect of medicine, and of death in the developed countries and is The European Society of Radiology. cardiology is at the forefront of this trend.2 one of the leading causes of disease burden He is a member of the editorial The unique nature of interventional in developing countries. In 2001, there review board of Cath Lab Digest. cardiology makes it an ideal target for the were 7.3 million deaths due to CHD development of AI-based technologies worldwide.9 An ST-segment elevation Chris is a Fellow of the Alliance of designed to improve real-time clinical myocardial infarction (STEMI) is a Cardiovascular Professionals, serves decision making, streamline workflow serious form of a heart attack in which Cardiac Credentialing International in the catheterization laboratory, and a coronary artery is completely blocked on the Board of Advisors, and was the standardize catheter-based procedures and a large part of myocardium is unable chair of the ASRT Cardiac & Vascular through advanced robotics.3 to receive blood. “ST-segment elevation” Interventional Chapter and served refers to a pattern that appears on an as a consultant to the ARRT Cardiac- The radiographer and cardiac electrocardiogram (EKG). This type of heart Interventional Exam Committee catheterization attack requires immediate, emergency and Continuing Qualification The cath lab has come a long way since revascularization which restores blood flow Requirements Advisory Committee. 1929 when Nobel Prize winner Werner through the artery. Forssmann performed the first cardiac Chris is the Cardiac Specialist/ catheterization on himself. Under local Time is muscle Invasive Cardiology Program Director anesthesia, Forssmann inserted a catheter Patient delay is a worldwide unsolved & Assistant Professor for Weber into a vein in his arm. Not knowing if the problem in STEMI. STEMI interventions State University in Ogden, Utah, USA. catheter might pierce a vein, he walked are unique in that the entire process is downstairs to the x-ray department where, enormously time-dependent and the
30 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 outcomes are linearly correlated with the the potential to empower a broader range AI at home system efficiency. Rapid diagnosis is critical of physicians to more accurately diagnose Patient data will not be collected solely in STEMI to facilitate timely intervention STEMI on ECG and reduce the inappropriate within the health care setting. Out-of- and requires immediate ECG evaluation. activation of the cath lab. This AI-based hospital cardiac arrest is a leading cause The 2013 ACCF/AHA and the 2012 and STEMI algorithm may contribute greatly of death worldwide.20 Rapid diagnosis and 2017 ESC guidelines stress the importance to the improvement of the current STEMI initiation of cardiopulmonary resuscitation of obtaining a 12-lead electrocardiogram system around the world. AI application in (CPR) is the cornerstone of therapy for (ECG) in a timely fashion (≤10 mins of ECG evaluation leads to higher accuracy victims of cardiac arrest. Yet a significant presentation).10,11,12 Door-to-balloon (D2B) and specificity, along with a faster number of cardiac arrest victims have no is a time measurement in emergency diagnosis. Faster diagnosis, in particular, chance of survival because they experience cardiac care specifically in the treatment translates into improved triage of patients an unwitnessed event, often in the privacy of STEMI. In patients with ST-segment– by decreasing unnecessary transfers of their own homes. The proliferation of elevation myocardial infarction, timely while improving outcomes and therefore, mobile sensors will allow physicians of the reperfusion therapy with D2B time <90 reducing costs.16 future to monitor, interpret, and respond minutes is recommended by the current to additional streams of biomedical data guidelines.13 The interval starts with AI and the treatment of acute myocardial collected remotely and automatically.21 the patient’s arrival in the emergency infarction Recent technology has provided access to a department and ends when a catheter The cardiac intervention has been the reliable means of obtaining an ECG reading guidewire crosses the culprit lesion in the main treatment for cardiovascular disease through a smartphone application.22 cardiac cath lab. Every minute of delay in in recent decades, including CHD and This technology could conceivably be treatment impacts mortality. The mortality acute coronary syndrome.17 Interventional widely disseminated and paired with ML of patients increases by 7.5% for every cardiology has traditionally been at the interpretation to rapidly triage patients 30 minutes that elapse before a patient leading edge of cardiovascular innovation. with STEMI. Expedited transfer to a with ST-segment elevation is recognized The past decade has seen an increasing facility capable of percutaneous coronary and treated.14 It has been reported that a emphasis on invasive intravascular intervention could then be facilitated in change in D2B from 15 to 180 min would imaging and physiology, exercise a timelier manner, potentially improving lead to an increase in in-hospital mortality hemodynamics, robotics, and a progressive outcomes. An under-appreciated diagnostic of 3% to 8.5% and an increase in 6-month blending of minimally invasive (as well element of cardiac arrest is the presence mortality from 10% to 20%.11 as invasive) cardiovascular surgical of agonal breathing, an audible biomarker, procedures.18 Soon, by using deep and brainstem reflex that arises in the AI and the diagnosis of acute learning, AI will have the ability to identify setting of severe hypoxia. ML algorithms myocardial infarction coronary atherosclerotic plaques more have been employed to analyze home Prehospital diagnosis of STEMI is of accurately than clinicians. Early research recordings via smart home speakers and major importance in reducing time to into the application of AI to assess phones to identify agonal breathing.23 treatment, in particular when patients can coronary arteries has included a recent Accurate detection of such recordings be transported directly to a facility with a study that validated the noninferiority could allow for the detection of arrest and cardiac catheterization lab, also known as a of AI to determine the physiologic activation of emergency response in the “cardiac cath lab.” However, obtaining 12- importance of coronary lesions and large percentage of arrests that occur lead electrocardiography (ECG) for STEMI recommendation for revascularization.19 unwitnessed at home. is cumbersome, and inefficient, relying on In addition, AI can also be used to analyze experts for confirmation. Considering the echocardiographic images, including The use of AI to assess cardiovascular critical role ECG plays in the early detection automatic measurement of the size of variables has gone mainstream as is of STEMI, an accurate warning system each chamber and assessment of left now being incorporated into wearable based on electrocardiogram (ECG) may be a ventricular function. Moreover, it can be devices. The most prominent example is solution for this problem, and AI may soon used to assess structural diseases, such the Apple Watch, backed by the American offer a path to improve its accuracy and as valvular disease, to help determine the Heart Association and U.S. Food and Drug cost-efficiency. An ECG monitor equipped classification and staging of the disease.17 Administration clearance. This wearable with a specific algorithm to automatically AI can be incorporated into the various device can detect atrial fibrillation.21 Via detect ST-segment elevation could specific components of both the STEMI photoplethysmography, the device traces theoretically help in pre-alerting STEMI process and the procedure. It can foster changes in blood flow and pulse. This and reducing patient delay. A potential large efficiency gains by simplifying each information is translated into a tachogram application of this AI-based ECG algorithm of the various components, be that at the that is processed via an AI-powered is its use by paramedics in ambulance individual patient, paramedic, ED, or the algorithm. If an irregular pulse is detected, and physicians from the referred hospital cath lab. Predictive diagnostics, therapeutic the algorithm can detect atrial fibrillation and clinics. It was reported that the strategy design, device selection, with a positive predictive value of 84%. The inappropriate activation of a cath lab for procedural optimization, and avoidance of user is notified to seek medical attention STEMI was around 20–25%15, which was complications are all areas in which the via an application installed on their iPhone. a great waste of medical resources. The application of AI is expected to make a This is a clear example of how technology application of this AI-based algorithm has rapid progression. can bring us closer to the patient and
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 31 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW improve screening, diagnosis, and at-home dose reduction and image optimization 8. Accessed July 20, 2020. patient monitoring.24 using algorithms are some elements of a www.world-heart-federation.org/wp- radiographer’s current role that could be content/uploads/2017/05/WEF_ Applications in Low and Middle-Income augmented with AI, but radiographers will Harvard_HE_GlobalEconomicBurden Countries still be responsible for keeping radiation NonCommunicable Diseases_2011.pdf Three-fourths of global deaths due to CHD doses as low as reasonably achievable. 9. Gaziano TA, Bitton A, Anand S, occurred in the low and middle-income Adoption of AI in medical imaging requires Abrahams-Gessel S, Murphy A. Growing countries.25 radiographers to adapt their imaging Epidemic of Coronary Heart Disease practices to ensure new technology is in Low- and Middle-Income Countries. In developing countries, rapid economic being implemented, used, and regulated Curr Probl Cardiol. 2010;35(2):72-115. transformation leads to environmental appropriately, based on high-quality doi:10.1016/j.cpcardiol.2009.10.002 changes and unhealthy lifestyles; in research evidence, maximizing benefits to 10. 2013 ACCF/AHA Guideline for addition, population aging may increase their patients.28 The successful application the Management of ST-Elevation cardiovascular risk factors and increase of AI will require close collaboration among Myocardial Infarction | Circulation. the incidence of CVD.26 CVD has placed a computer scientists, clinical investigators, Accessed July 19, 2020. heavy burden on patients and society as a cardiologists, radiographers to identify the www.ahajournals.org/doi/full/10.1161/ whole. Therefore, it is necessary to provide most relevant problems to be solved and cir.0b013e3182742cf6 strategies for improving the diagnosis and the best approach and data sources to do it. 11. Ibanez B, James S, Agewall S, et al. treatment of CVD in the future. AI may have 2017 ESC Guidelines for the the potential to solve this problem. References management of acute myocardial 1. Artificial Intelligence: Refining infarction in patients presenting with Medical AI technology not only could STEMI Interventions. Cath Lab ST-segment elevation. The Task improve physicians’ efficiency and quality of Digest. Accessed July 9, 2020. Force for the management of acute medical services, but other health workers www.cathlabdigest.com/content/ myocardial infarction in patients could also be trained to use this technique artificial-intelligence-refining-stemi- presenting with ST-segment elevation to compensate for the lack of physicians, interventions of the European Society of Cardiology thereby improving the availability of 2. Lopez-Jimenez F, Attia Z, Arruda- (ESC). Eur Heart J. 2018;39(2):119-177. healthcare access and medical service Olson AM, et al. Artificial Intelligence doi:10.1093/eurheartj/ehx393 quality. Recently, healthcare in China’s in Cardiology: Present and Future. 12. Steg PG, James SK, Atar D, et al. rural areas has been benefiting from Mayo Clin Proc. 2020;95(5):1015-1039. ESC Guidelines for the management medical AI technology. According to a doi:10.1016/j.mayocp.2020.01.038 of acute myocardial infarction in news report from the South China Morning 3. Sardar P, Abbott JD, Kundu A, Aronow patients presenting with ST-segment Post, a portable all-in-one diagnostic HD, Granada JF, Giri J. Impact of elevationThe Task Force on the station (weighing 11 pounds), which can Artificial Intelligence on Interventional management of ST-segment elevation run 11 tests, including blood pressure, Cardiology: From Decision-Making Aid acute myocardial infarction of the electrocardiographs, and routine urine and to Advanced Interventional Procedure European Society of Cardiology (ESC). blood analyses, has been used in village Assistance. JACC Cardiovasc Interv. Eur Heart J. 2012;33(20):2569-2619. healthcare settings. This device, which 2019;12(14):1293-1303. doi:10.1016/j. doi:10.1093/eurheartj/ehs215 was developed by an internet healthcare jcin.2019.04.048 13. Park Jonghanne, Choi Ki Hong, Lee Joo company supported by a national rural 4. Agrawal K. The First Catheterization. Myung, et al. Prognostic Implications healthcare program, can automatically The Hospitalist. Accessed July 22, 2020. of Door-to-Balloon Time and Onset- upload results and medical records to an www.the-hospitalist.org/hospitalist/ to-Door Time on Mortality in Patients online data analysis system and generate article/123249/first-catheterization With ST-Segment–Elevation Myocardial a diagnosis for village health workers to 5. Belenkov IN, Samko AN, Batyraliev TA, Infarction Treated With Primary review and reference.27 Pershukov IV. [Coronary angioplasty: Percutaneous Coronary Intervention. view through 30 years]. Kardiologiia. J Am Heart Assoc. 2019;8(9):e012188. The role of the radiographer 2007;47(9):4-14. doi:10.1161/JAHA.119.012188 The cath lab-based radiographer has been 6. Roth GA, Abate D, Abate KH, et 14. De Luca Giuseppe, Suryapranata keeping pace with the rapidly evolving field al. Global, regional, and national Harry, Ottervanger Jan Paul, Antman of cardiology since first catheterization age-sex-specific mortality for 282 Elliott M. Time Delay to Treatment and 1929. There have been major advancements causes of death in 195 countries Mortality in Primary Angioplasty for in the use of AI in recent years in nearly all and territories, 1980–2017: a Acute Myocardial Infarction. Circulation. areas of cardiology, with some concrete systematic analysis for the Global 2004;109(10):1223-1225. doi:10.1161/01. advances in ECG analysis, automatic Burden of Disease Study 2017. The CIR.0000121424.76486.20 interpretation of imaging studies, and risk Lancet. 2018;392(10159):1736-1788. 15. Zhao Y, Xiong J, Hou Y, et al. prediction. However, AI is susceptible to doi:10.1016/S0140-6736(18)32203-7 Early detection of ST-segment major errors in interpretation, validity, 7. Organization WH. Global Status Report elevated myocardial infarction by and generalizability, raising safety, and on Noncommunicable Diseases 2014. artificial intelligence with 12-lead ethical concerns. The opportunity for World Health Organization; 2014. electrocardiogram. Int J Cardiol.
32 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 Published online May 3, 2020. pmc2864143?pdf=render doi:10.1016/j.ijcard.2020.04.089 26. Wu Y, Benjamin EJ, MacMahon 16. A Primer on Artificial Intelligence (AI). S. Prevention and Control Plus: Is Your Job on the Line? Cath Lab of Cardiovascular Disease Digest. Accessed July 22, 2020. https:// in the Rapidly Changing www.cathlabdigest.com/article/Primer- Economy of China. Circulation. Artificial-Intelligence-AI-Plus-Your- 2016;133(24):2545-2560. doi:10.1161/ Job-Line CIRCULATIONAHA.115.008728 17. Yan Y, Zhang J-W, Zang G-Y, Pu J. The 27. Guo J, Li B. The Application of Medical primary use of artificial intelligence Artificial Intelligence Technology in in cardiovascular diseases: what Rural Areas of Developing Countries. kind of potential role does artificial Health Equity. 2018;2(1):174-181. intelligence play in future medicine? doi:10.1089/heq.2018.0037 J Geriatr Cardiol JGC. 2019;16(8):585- 28. The European Federation of 591. doi:10.11909/j.issn.1671- Radiographer Societies. Artificial 5411.2019.08.010 Intelligence and the Radiographer/ 18. Lopez-Jimenez F, Attia Z, Arruda- Radiological Technologist Profession: Olson AM, et al. Artificial Intelligence A joint statement of the International in Cardiology: Present and Future. Society of Radiographers and Mayo Clin Proc. 2020;95(5):1015-1039. Radiological Technologists and the doi:10.1016/j.mayocp.2020.01.038 European Federation of Radiographer 19. Davies J. CEREBRIA-1: machine Societies. Radiography. 2020;26(2):93- learning vs expert human opinion to 95. doi:10.1016/j.radi.2020.03.007 determine physiologically optimized coronary revascularization strategies. Presented at the: September 24, 2018; TCT 2018. 20. Myat A, Song K-J, Rea T. Out-of-hospital cardiac arrest: current concepts. The Lancet. 2018;391(10124):970-979. doi:10.1016/S0140-6736(18)30472-0 21. Johnson KW, Torres Soto J, Glicksberg BS, et al. Artificial Intelligence in Cardiology. J Am Coll Cardiol. 2018;71(23):2668-2679. doi:10.1016/j. jacc.2018.03.521 22. Barbagelata A, Bethea CF, Severance HW, et al. Smartphone ECG for evaluation of ST-segment elevation myocardial infarction (STEMI): Design of the ST LEUIS International Multicenter Study. J Electrocardiol. 2018;51(2):260-264. doi:10.1016/j. jelectrocard.2017.10.011 23. Chan J, Rea T, Gollakota S, Sunshine JE. Contactless cardiac arrest detection using smart devices. Npj Digit Med. 2019;2(1):1-8. doi:10.1038/s41746-019- 0128-7 24. Garvey JL, Zegre-Hemsey J, Gregg R, Studnek JR. Electrocardiographic diagnosis of ST segment elevation myocardial infarction: An evaluation of three automated interpretation algorithms. J Electrocardiol. 2016;49(5):728-732. doi:10.1016/j. jelectrocard.2016.04.010 25. Accepted Version. Accessed July 20, 2020. https://europepmc.org/articles/
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 33 ISRRT SOCIETIES – MEMBER VIEWS
ASRT 2019 Artificial Intelligence survey results regularly operate. Self-driving, guidance, and USA THE American Society of Radiologic Technologists has automatic positioning were the most recognized AI and over 157,000 members and is committed to advancing ML features across modalities, except for radiation and elevating the medical imaging and radiation therapy therapy. Anatomical detection also was quite common. professions to enhance the quality and safety of patient Some examples of anatomical detection include care. We do this by staying abreast of current and future vessel detection in vascular intervention, structure technological changes and seeking to understand contouring for radiation therapy, and automatic image professional acceptance of coming changes. To better reconstruction for magnetic resonance imaging. understand the advancement of Artificial Intelligence Nearly 55% of respondents noted that automated in Medical Imaging and Radiation Therapy, the ASRT postprocessing is common practice in their department. conducted a survey of its members. The 2019 Artificial Respondents were asked how frequently they use AI and Intelligence Survey was sent via email to 20,000 ASRT ML features of equipment, and over 18% stated they “use members in August 2019. A total of 416 participants them all the time”, 34.5% “use them most of the time”, completed the questionnaire, for a 2.1% response rate. 20.9% “use them some of the time”, 10.4% “rarely use This sample size yielded a margin of error of ± 4.8% them”, and 15.9% “never use them”. at the 95% confidence level. Most participants work at a hospital (58.6%), while many others work at an The areas respondents feel AI and ML will most benefit imaging center (10.2%) or a large clinic (8.5%) with the practice are motion artifact recognition (72.2%), radiation majority in an urban setting (44.9%) or suburban setting exposure control (71.4%), and quality of scans and (38.8%). The remaining 16% work in a rural setting. The treatments (70.4%). Respondents foresee a neutral respondents are staff technologists or therapists (69.2%), effect on throughput (53.2%), creativity (52.9%), and chief technologists (9.7%), and supervisors or assistant responsible implementation (52.4%), as well as negative chief technologists (5.8%). Most work full time (82.3%) effects on patient interaction (30.8%) and expense and the average age of respondents is 42.9 years. (25.6%). A majority (53.5%) believes AI will not affect the technologist’s scope of practice. Nearly 31% of those age Artificial Intelligence (AI) is broadly defined as the 18 to 24, the largest age cohort, believe their role will simulation of human-like intelligence by machines. be expanded, 33% of respondents age 34 to 43 believe The challenge of AI is designing computer systems to AI will reduce their professional role, and 80% age 63 or perform tasks that require human intelligence, older anticipate no change, likely due to retirement. including visual perception, speech recognition, and Lastly, the survey evaluated how technologists perceive decision-making. Machine Learning (ML) is one AI and ML will affect staffing and specialized roles. approach to artificial intelligence that gives computers Almost 63% of respondents across all age groups the ability to learn without being programmed. Most believe staffing will not be affected. Those age 34 to respondents agreed that these definitions matched 43 are most concerned about a staffing reduction, and their understanding of these concepts (88.6% AI and respondents 63 and older are the least concerned. The 72.4% ML). 54 to 62 age group somewhat believes additional staffing will be necessary. Half of the 18 to 24 group believes Survey results suggest that respondents, who primarily AI will lead to the development of new roles within the work in radiography (54.7%), computed tomography profession, while the 25 to 33 demographic is most likely (20.7%), and mammography (13.3%), are largely to anticipate a reduced role. comfortable with technology and frequently use it in their everyday life. Though most respondents are broadly Technologists are mostly comfortable with technology familiar with the concepts of AI and ML, they have mixed in everyday life, use it daily at work and are cautiously experiences with AI features on imaging equipment. optimistic about the incorporation of AI and ML Regardless, many are confident that those features technologies in their departments. Most respondents function correctly and provide trustworthy results. somewhat trust AI algorithms or trust them a great deal. One area of concern was the lack of standardized The results are encouraging and demonstrate current processes for resolving discrepancies between use of AI and a general acceptance of change. The machine-recommended procedures and technologist majority feel that AI will have a positive impact on safety judgment. However, the majority feel that AI will have and quality, with a mild concern that “human” aspects of a positive impact on safety and quality, with a mild the profession such as patient interaction and creativity concern that “human” aspects of the profession such could suffer. While it is generally believed that AI could as patient interaction and creativity could suffer. There elevate patient care, we must assure that the human was no consensus that AI would adversely affect their aspect of imaging remains. professional prospects. Melissa B. Jackowski, Ed.D., R.T.(R)(M) AI and ML features of equipment were analyzed for Michael Jennings, M.A. several modalities, and results were cross tabulated John Culbertson, M.A., M.Ed. to factor out equipment that technologists do not Craig St. George, M.S., R.T.(R)(VI)
34 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 AI In Cambodia tell when patients are deteriorating It can save costs Cambodia Elevating Patient Care with Artificial Intelligence for both the hospitals and patients and precision of “Radiographers are essential in elevating patient care machine leaning can detect diseases such as cancer with artificial Intelligence” disease sooner and can get the right way treatments and saving patients life, best medical service, accurate Objective diagnosis and trust patients information before the As a post-conflict nation, emerging from decades of patient needs hospitalization, however AI could be takes civil war, Cambodia faces limitations on its country- data from patients to help improve patient outcomes, It wide infrastructure, but gaps in nascent systems can improves reliability, predictability, and consistency with present a unique opportunity to innovate with digital quality and patient safety and will increase efficiency, it technologies such as AI (artificial Intelligence). It improves human performance, it is used as a decision was assumed that economic development required augmentation tool, while it can’t replace doctor and industrial infrastructure. But the fourth Industrial nurses, it can make them more effective, efficient and revolution offers the possibility for any country to happier on the job which increases confidence as well simply leapfrog infrastructure development. As as reduces stress and anxiety. Cambodia is a developing country, so the Artificial intelligence in medical and healthcare has been Conclusion a particularly under considering for hospital and In the future we looking for deployment of AI represents organization, I have seen a small number of AI. a unique challenge for every nation and nowhere more Venture companies attend in Cambodia, its upcoming so than in the improvement of healthcare. Cambodia wave of trends is coming a few years before which faces unique challenges and opportunities anchored affected with the price. For our experience what AI in its unique social and technological infrastructure can perform in Cambodia is gathering of data through and will serve as a model for technology deployment in patient interviews and tests processing and analyzing tackling common global problems develop a medical results using multiple sources of data to come to education system to create a workforce competent in an accurate diagnosis determining an appropriate the utilization, evaluation, and improvement of AI and treatment method often presenting options preparing to deploy thoughtful multidisciplinary and multilateral and administering the chosen treatment method efforts to identify and disseminate best practices patient monitoring, an additional AI methods excel at from patient-care to system-wide management. automatically recognizing complex patterns in imaging Through such an approach, it is uniquely situated to data and providing quantitative, rather than qualitative, demonstrate to the world a well-designed roadmap of assessments of radiographic characteristics and a lot an AI-driven future of healthcare. of advantages to use A in work place. However, if we could has are many applications of AI within radiology Sophea Ly that are beyond image interpretation and may even be implemented much earlier in actual practice enhance radiology through improving imaging appropriateness and utilization, patient scheduling, exam protocoling, image quality, scanner efficiency, radiation exposure, radiologist workflow and reporting, patient follow-up and safety, billing, research and education, and more to improve, ultimately, patient care, helpful for doctor finding the cause of disease and provide specific treatment to patients, help decreasing patient visited abroad to get medical service from there.
Methods Presently collaborations between hospitals and industry, Cambodia has a several abroad company were investments in medical field that used AI for provides service to local patients and traveling patients such as treatments and diagnostic criteria by offering international standard medical services in the capital, and our expected is look into how much AI can be attend in Cambodia in the future even currently AI system was used in several hospital and only private hospital or diagnosis center, because the AI price were a problems even this technology is used in health care is very helpful for doctors recognize diagnoses and
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 35 ISRRT SOCIETIES – MEMBER VIEWS
Artificial Intelligence in diagnostic radiography the strengths of the human factor in the role and Ireland effectiveness of the profession means we need to FROM the invention of the earliest diagnostic imaging keep the human in any development in AI. We need to devices using Roentgen inspired X-ray tubes in the set aside dividing lines and think of it not as Artificial early 1900s to the use of non-ionising radiation Intelligence but rather Augmented Intelligence7, with technologies of today such as ultrasound and us as part of the equation (or algorithm even) and this magnetic resonance imaging1 the profession of the is the next adaptation of the profession. We should radiographer has required constant adaptation in the welcome improved decision support tools that enhance face of rapid technological developments. Not only our performance and the care of our patients and we adaptation to technology were required but constant should also to continue to monitor the examinations role expansion including the management of risks of to ensure sacrosanct principles such as ALARA are using imaging technologies that were missed by the maintained. We can use the improved patient contact early pioneers. time afforded us by this technology (just as we did with the replacement of film-screen by digital images) to Radiographers have been using “machine intelligence” focus more on our patients making sure they are better for quite some time in the form of Automatic informed, better prepared and ultimately provided Positioning of equipment based on procedure in better healthcare. interventional radiology suites, Automatic Exposure Control (AEC) of X-ray imaging systems and In this respect, radiographers have not only an ongoing segmentation tools or computer aided detection role in adopting this AI technology, but also in being in image post-processing to name but a few. The part of the development, testing, and safe, supervised level of automation continues to grow year on year use of this technology.8 with additional tools being provided to assist the radiographer/RT in the performance of their duties. Dr John Stowe, It is therefore of no surprise that we will see the Assistant Professor Radiography and Diagnostic evolution of these assistance tools with the growth Imaging, School of Medicine, UCD, of Artificial Intelligence (AI), Machine Learning (ML) Dublin, Ireland for the IIRT and Deep Learning (DL) technologies. Many papers consider “Is AI a threat to our profession”2 and we need to be aware of the implications for the radiography Artificial Intelligence in radiation therapy profession. Radiation Therapists (RTTs) have a well-recognised thirst for progress, and a willingness to embrace Key to this is understanding what AI/ML/DL can and advances in technology. With these inherent cannot do. It is, when well developed, an incredible professional qualities, it comes as no surprise that system for rapidly determining patterns or sequences the integration of artificial intelligence (AI) into clinical that can be difficult for humans to understand in a practice has been explored across a number of areas timely fashion. It can assist with many arduous duties of radiation oncology. This brief overview will highlight like radiation dose control but do not let the word the some of the current uses of AI throughout the chain intelligence in AI fool you into thinking it is all knowing. of radiation therapy and will demonstrate the potential Even with AEC we still have seen over exposure impact on clinical practice. instances that “machine intelligence” did not prevent Richwine3. We have seen from our professional counterparts in diagnostic radiography, that there are many benefits Machine learning loves data that contains repeatable, of AI related to image acquisition. With this application even if difficult to identify, patterns which occur in a lot extended to radiation therapy, AI methods have enabled of real-world digital data. But what about emotional the generation of synthetic CT images based on MRI data, mis-understanding the meaning of a simple smile data. Through the use of complex neural networks can have serious consequences for action2. AI efforts (CNN), the high-resolution synthetic CT images are being made to recognise meaning in speech, as generated are of sufficient quality for radiotherapy opposed to word/phrase recognition but they are not planning purposes9. This application is promising exhibiting true understanding of speech and this can as it negates the need for a potentially unnecessary result in problems4. AI suffers from bias5, lacking the planning CT to be acquired, paving the way for MRI potential to adapt given the constraints of its localised only simulation, thereby reducing dose to the patient area of operation (or domain) and the data available and improving workflow. Furthermore, as Rattan et al., to it (6). Human communication is complicated and point out, the combination of MRI and CNN generated patients require more than “the illusion of intelligence synthetic CTs reduces image fusion uncertainties which or humanness”4. could potentially impact on delineation and treatment planning downstream9. AI is part of radiography, quite literally. Recognising
36 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 Image segmentation is arguably one of the most professional roles in ways we never imagined. time-consuming aspects of radiation therapy and here we are reaping the benefits of AI that has Elizabeth Forde, pushed us beyond the current standard practice of Assistant Professor, School of Radiation Therapy, atlas-based auto-segmentation. Multiple studies Trinity College Dublin, Ireland for the IIRT in the literature have demonstrated the significant reduction in contouring time, for example Lustberg References et al., reported a median time saving of nearly 80% 1. Scatliff JH, Morris PJ. From Röntgen to Magnetic through the use of deep learning methods.10 With Resonance Imaging. North Carolina Medical the shift towards online adaptive radiotherapy (ART) Journal. 2014;75(2):111. where delineation, optimisation and dose calculation 2. Moehrle A. “Radiology” Is Going Away . . . and That’s occur whilst the patient is on the treatment couch, this Okay: Titles Change, A Profession Evolves. Journal resultant reduction in burdensome contouring tasks of the American College of Radiology. 2018;15(3, is particularly valuable. Spieler et al., illustrated this Part B):499-500. particular application with their recent online ART 3. U.S. probing more cases of CT radiation study for patients with pancreatic cancer; a site which overexposure, (2009). is particularly challenging to contour.11 While most of 4. Neff G, Nagy P. Talking to Bots: Symbiotic Agency the studies in AI driven segmentation acknowledge a and the Case of Tay. International Journal of consistent reduction in time, there have been mixed Communication. 2016;10:4915-31. results reported with respect to contouring accuracy. 5. Rhue L. Racial Influence on Automated Perceptions And whilst more work is yet to be done in this field, of Emotions (November 9, 2018). 2018. sub-optimal preliminary results are predominantly 6. Ribeiro MT, Singh S, Guestrin C. “Why Should I Trust caused by inconsistencies in the manual contours found You?”: Explaining the Predictions of Any Classifier. within the model training datasets.10 HLT-NAACL Demos 20162016. 7. Crigger E, Hhoury C. Making Policy on Augmented Knowledge-based planning is not a new concept in Intelligence in Health Care. AMA J Ethics. radiotherapy and has resulted in improved efficiency 2019;21(2). in plan generation, as well as improved consistency 8. ISRRT/EFRS. Artificial Intelligence and the in the quality of plans between individual planners. Radiographer/Radiological Technologist Profession: The integration of AI in this regard was the next A joint statement of the International Society of natural step and has also garnered interest in the Radiographers and Radiological Technologists and literature. One such example of this novel application the European Federation of Radiographer Societies. was demonstrated by Fan et al.12 Here, using deep Radiography. 2020;26(2):93-5 learning algorithms, a model was trained based 9. Rattan R, et al. Artificial intelligence in oncology, its on 270 nasopharyngeal IMRT plans to predict the scope and future prospects with specific reference dose distribution of subsequent patients. Following to radiation oncology. BJR Open 2019; 1: 20180031. validation, their model was able to not just predict 10. Lustberg T, et al. Clinical evaluation of atlas and an achievable distribution, but also converted this deep learning based automatic contouring for lung information into usable optimisation parameters cancer. Radiother Oncol 2018; 126(no. 2): ISSN: to automatically generate a clinically acceptable 1879-0887): 312: 710.1016/j. radonc.2017.11.012.. plan. By selecting a disease site that is notoriously doi: https://doi.org/10. 1016/j.radonc.2017.11.012 time consuming to plan, Fan et al., have clearly 11. Spieler B et al., Automatic segmentation of demonstrating the promising scope for AI in treatment abdominal anatomy by artificial intelligence (AI) in planning.12 Furthermore, head and neck patients adaptice radiotherapy of pancreatic cancer. IJROBP. often require unscheduled ART due to weight loss 2019; 105(1):E130-131 and tumour shrinkage during treatment. The role 12. Fan J et al., Automatic treatment planning based on of AI to expedite the laborious planning process three-dimensional dose distribution predicted from prevents potential breaks in treatment, which from a deep learning technique. Medical Physics. 2019; radiobiological perspective is particularly attractive for 46(1):370-381 this specific patient group. 13. Jiang S. The impact of AI in radiation oncology. AAPM 2018. Available from: www.itnonline.com/ Transferring AI methodologies from a confined videos/video-impact-artificial-intelligence- research setting into the clinical environment will have radiation-therapy. Cited July 1st 2020. a tangible impact on department workflow and patient care. The scope to reduce any existing disparities in quality of care by exploiting AI to allow for knowledge transfer between staff and across institutions is particularly exciting.13 Likewise, is the scope for RTTs to advance their practice even further and adapt our
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 37 ISRRT SOCIETIES – MEMBER VIEWS
Usefulness of AI (Artificial intelligence) in Japan length product): 402.9mGy-cm, Effective dose: 5.64mSv) Japan and low dose (SD17, CTDIvol: 4.4mGy, DLP: 139.3mGy- AI (Artificial intelligence) is now being used in various cm, Effective dose: 1.95mSv) are shown below. The situations in Japan. In the medical field, it has been effective dose was calculated using adult chest k-factor picked up as the latest topic such as doctor’s support (0.014 mSv-mGy-1-cm-1) of ICRP Pub.102. In the past, diagnosis assistance1,2 and patient assistance in when trying to reconstruct a low-dose image to ensure rehabilitation. Regarding the support diagnosis the conventional image quality, it took a long time to of doctors, it is a so-called CAD (computer-aided calculate and was difficult to use clinically. Although, it diagnosis), which started in 1960 and has a history of is possible to speed up the process by using machine over 20 years. This CAD is now called AI-CAD, which learning in the calculation process and make it easier uses AI for machine learning and deep learning. The to use clinically, and it is possible to reduce radiation accuracy is improved, and further evolution is expected dose by about 35% without degrading the image quality in the future. In the field of radiological diagnosis, it (under figure). However, the image reconstruction and has begun to be adopted in most fields such as general software using these AIs are black boxes, and there imaging, CT (computed tomography)3, MRI (magnetic are few detailed explanations from manufacturers. resonance imaging)4, nuclear medicine5, radiotherapy.6 Therefore, “Is the image and data output after scanning In particular, it has become possible to reduce really correct?” “Can the radiation dose be reduced?” the radiation dose to patients by using AI to obtain It is a problem point of AI that it is not known at conventional image quality with a low-dose X-ray. It has present how to output images and data and it is a been able to perform with high accuracy examination future task to clarify it. It is considered necessary to than so far by incorporating AI into software. In the CT evaluate the accuracy of the data and the information field, if AI is introduced into the dual energy technology output by AI. In the future, it is thought that our and accuracy is improved, it is expected that the radiological technologist, will be able to provide a better examination will be able to obtain more information examination with minimally invasiveness for patients by such as functions from the examination that was elucidating and using it. mainly based on morphological information. There are machine learning, neural network, and deep learning in Reference AI. Neural networks are a method of machine learning 1. Yasaka, K., Akai, H., Abe, O., Kiryu, S.: Deep Learning that mimics the neural network of the human brain. with Convolution Neural Network for Differentiation Deep learning is being learned and advanced by itself. of Liver Masses at Dynamic Contrast-enhanced CT; Although these are collectively called AI, they are A Preliminary Study. Radiology, 286(3), 887-896, applied to the medical field by taking advantage of their 2018. respective characteristics. 2. Nishio, M., Sugiyama, O., Yakami, M., et al.: Computer-aided diagnosis of lung nodules I am going to introduce some parts of CT in my classification between benign nodule, primary lung specialty. In my current research, AI is not used for cancer, and metastatic lung cancer at different reconstruction, but it is used in the calculation process image size using deep convolutional neural network to speed up calculation. The images of the coronal with transfer learning. Plos ONE, 13(7), 2018: section of the chest phantom with simulated nodule e0200721. taken with conventional dose (SD(standard deviation) 3. Motonori, A., Yuko, N., Toru, H.: Deep learning 10, CTDIvol(volume CT dose index): 12.7mGy, DLP(dose reconstruction improves image quality of abdominal
Conventional dose image. Low dose image.
38 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 ultra-high-resolution CT. European Radiology, 29, 6. Tomori, S., Kadoya, N., et al.: A deep learning-based 6163–6171, 2019. prediction model for gamma evaluation in patient- 4. Takenaga, T.,Hanaoka, S., Abe, O., et al.: Four- specific quality assurance.Medical Physics, 45(9), dimensional fully convolutional residual network- 4055-4065, 2018. based liver segmentation in Gd-EOB-DTPA- 7. ICRP, 2007. Managing Patient Dose in Multi-Detector enhanced MRI. International journal of Computer Computed Tomography (MDCT). ICRP Publication Assisted Radiology and Surgery, 14(8), 1259-1266, 102. Ann. ICRP 37 (1). 2019. 5. Hashimoto, F., Ohba, H., Ote, K., Teramoto, A., Hideki Shibata Tsukada, H.: Dynamic PET image Denoising Using Toyota Kosei Hospital deep Convolutional Neural Networks Without Prior Training Datasets. IEEE Access, 7, 96594-96603, 2019.
Elevating patient care with Artificial Intelligence: A new model that was developed, called the LOAD Nigeria Radiographers are essential in elevating patient model that can assist in identifying factors that can care with Artificial Intelligence: cause problems when we introduce new functionality Perspectives of Nigerian radiographers like machine learning (ML) workflows to complex ecosystems. The LOAD model stands for; Landscape; Introduction Organization; Action; Data. The model describes RADIOGRAPHERS worldwide have been at the fore four dimensions that can affect the adoption of new front of the use of modern digital technology in medical digitalization innovations such as AI tools. The rationale imaging and radiation therapy for improved patient of the model is to provide some structure on identifying care. This is because they have maintained the human socio-technological factors when implementing new AI touch between emerging technologies and the patient tools to be embedded in complex environments. Some over the years. It is against this background that the of this tools includes image segmentation and image Association of Radiographers of Nigeria (ARN) throws its interpretation in radiotherapy. By interrogating the four weight behind the position of the International Society dimensions of the LOAD model, domain experts are of Radiographers and Radiologic Technologists (ISRRT) covering factors, stemming not only from technical, which is the voice of Radiographers worldwide that; but also from organizational and human sources. In Radiographers are Essential in Elevating Patient Care the LOAD model, the data remains the focal point of with Artificial Intelligence. investigation. All the other dimensions are analyzed from the data perspective. While the development and deployment of Artificial Intelligence in health care in most low and middle The LOAD model was a concept developed based on income countries, Nigeria inclusive is yet to fully take research on factors contributing towards the failure root, as proactive professionals, we recognize the to realize the planned benefits of new technologies, fact that the future is now. Hence an urgent need for IT systems in the NHS in the UK. The study analyzed adequate preparation. We therefore urge the ISRRT. 18 case studies by staff working for the NHS. The case studies describe factors responsible for failure Our Position or success of recent developments due to a mixture It is important for us as radiographers to ask ourselves technical and human factor with human factors being by this important question, whether our data landscape far the most dominant. The most common cause of IT is ready to accommodate artificial intelligence. This is failure in these case studies is related to people and their important because radiographers generate enormous interactions. amount of data in the course of imaging patients during medical imaging and radiation therapy procedures every Hence, to investigate if our current data landscape is day. The modern health care ecosystem which also ready for new machine learning techniques and tools, includes radiography is ever dynamic, complex and data we have to investigate not only the technical but also driven. Data is considered the most valuable asset of the human and organizational factors that can pose any the 21st century. Data if used in Artificial Intelligence threat to the new project. (AI) workflow can transform device, give back the gift of time to clinicians and maximize patient care especially in The data dimension of the model describes factors disease recognition and interpretation. introduced by the data needed to be used in the new
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 39 ISRRT SOCIETIES – MEMBER VIEWS
system or functionality. Data entry is a time consuming disillusionment and another AI winter, and/or further process and when done by a less experienced staff could exacerbate existing health and technology driven introduces error which could significantly reduce the disparities. Radiographers are relevant stakeholders quality of information. in elevating patient care with artificial intelligence in health care. People and systems interacting with data to accomplish some value are referred to as actors. For example data 2. Considering the fact that AI tools are being entered into a system by secretarial staff can contain implemented in an environment of low regulation and error if the information requires medical knowledge legislation particularly in low resource settings, it is vocabulary that the staff lacks. We refer to this as the important to ensure that human-centered AI tools risk of a class of grammers. include accepting that human override is important for developing user-trust because the public has an Moving onto the organization dimension, sharing outside understandably low tolerance for machine error. immediate organizational unit can result in a number of administrative costs, such as reaching and complying 3. The near term dialogue around AI in health care with data sharing agreements, as well as complying should focus on promoting, developing, and with wider information governance requirements. The evaluating tools that support humans rather than data landscape finally describes the environment in replace it with full automation systems reliance which the other three components; organization, actors on data. and data interact. AI Data are critical for developing any AI algorithm. Preparing the radiography workforce that will deliver Without data, the underlying characteristics of the the digital future by elevating patient care with artificial process and outcomes are unknown. Paucity of intelligence entails a critical review of important aspects quality data in health care has been a challenge in and future practice as radiographers; how technological health care for decades, but key trends (such as and other developments is likely to change the roles of commodity wearable technologies) in this domain in radiographers over the next two decades to ensure safer, the last decade have transformed health care into a more productive, more personal care for patients; the heterogeneous data-rich environment (Schulte and implications of these changes are for the skills required Fry, 2019). by radiographers filling these roles; the consequences for selection, curricula, education training, development Databias and lifelong learning of qualified radiographers and Selecting an appropriate AI training data source is those in training are key competences that AI intends to critical because training data influences the output address in our national policy. observations, interpretations, and recommendations. If the training data are systematically biased due, AI has been identified as one of the means of addressing for example, underrepresentation of individuals of the health care challenges of the 21st century. There a particular gender, race, age or sexual orientation, must be mechanisms in place to ensure advanced those biases will be modeled, propagated, and scaled technologies do not dehumanize care. While automation in the resulting algorithm. The same is true for human will improve efficiency, it should not replace human biases (intentional or not operating in the environment, interaction. workflow, and outcomes from which the data were collected. Important points to support deployment of AI in Radiography and Radiation Therapy Practice: Recommendations 1. The emergence of artificial intelligence(AI) as a tool • Radiographers as well as clinicians must consider for better health care offers uncommon opportunities the potential impact of the patient to radiographer to improve patient and clinical team outcomes, relationship in the age of AI-enabled health care. reduce costs, and impact population health. Examples • To achieve high performance ,radiographers include but are not limited to automation; providing should carefully consider strategies for advancing patient, family (friends and family unpaid care givers), compassionate, high quality care within the digital health care professionals information synthesis, ecosystem . recommendations and visualization of information for To participate in the shaping of AI practices and shared decision making. AI organizations, professionals must have the competencies and the capabilities to participate in Nigerian radiographers while being cognizant of the the development and shaping of the future of their numerous promising examples of AI applications discipline and practice. in health care as seen in literature both within and • For Radiographers and other health professionals to beyond radiography believe that it is important to stay relevant, they must ensure that their trainees proceed with caution, else we may end up with user have the required competencies and that their active
40 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 members evolve their capabilities to adapt their intelligence-enabled health care practice: Rewiring practices in an AI-enabled environment. health care professions for better care. Journal of • How vast are radiographers to harness the medical imaging and radiation sciences .50,8-14 knowledge, skills, and attitudes to effective usage of AI tools to improve outcomes for patients and Flavious Nkubli their communities Lecturer and Honorary clinical Radiographer, Department of Radiography University of Maiduguri/ Reference Department of Radiology University of Maiduguri Wiljer D, Hakim Z (2019). Developing an artificial Teaching Hospital
Society of Radiographers of South Africa: delegates interacted with the speakers throughout the South 1st Virtual Symposium - presentations and results from the polls and surveys ‘Radiographers are essential in elevating patient care indicate that delegates not only found the symposium to Africa with artificial intelligence’ be of value but suggest that more needs to be covered related to AI and the radiographer. Well done to the ARTIFICIAL intelligence (AI) has been predicted as ISRRT for choosing this theme for WRD 2020. taking over radiology. These predictions may be ‘attention grabbing’ however it is highly unlikely that a We are fortunate and grateful to our sponsors who machine will replace us anytime soon. AI will augment assisted in making this symposium a success. but not replace and take over the profession. Please check www.sorsavirtual.co.za for more The Society of Radiographers of South Africa (SORSA) information. identified the important role of AI in our profession. We are proud to announce that we hosted our first Watch this space for an upcoming symposia - www. virtual symposium on September 12, 2020. The theme sorsavirtual.co.za and the Society of Radiographers of of the symposium was aligned with the ISRRT World SA_NC facebook page. Radiography 2020 theme: Radiographers are essential in elevating patient care with artificial intelligence. Fozy Peer The symposium was complimentary to all delegates SORSA: Public Liaison Officer nationally and internationally, but registration was necessary. It was accredited for 2 ethics and 1 general CEU (continuing education unit) and delegates had to indicate if they required CPD certificate. Attendance was monitored per session. Registrants may access the pre-recorded presentations via the zoom platform. The presentations are available for 90 days after the symposium.
The speakers were of exceptionally high calibre, each one being an expert in his/her respective field. They covered subjects related to the ethics and the medico-legal aspects of AI. Presentations pertaining to the education required for AI, and programmes related to scheduling in AI for the ease of the radiographer were also covered.
There were just under 500 registered delegates. The
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 41 ISRRT SOCIETIES – MEMBER VIEWS
Elevating patient care with Artificial Intelligence: appreciate the true disruptive potential of AI and Singapore Radiographers are essential in elevating patient exploit the benefits that this technology brings. care with Artificial Intelligence The challenge for the profession is to remain adaptable ARTIFICIAL Intelligence (AI) has made significant to technological innovation. To do so necessitates that advances in radiological and radiation oncology. the medical radiation professions of both diagnostic AI is already a synonym in radiology for image and radiation therapy tracks be engaged and shape interpretation and decision-making, allowing it to the conversation on AI to inform our future practice. complement the clinical practice for safe, quality care. One way is through aligning our expectation of AI by Progress in machine learning, deep learning, neural understanding what AI can and cannot do, and how, networks, natural language processing, and quality by harnessing AI, can we deliver excellent patient care. datasets, will undoubtedly catapult the possibility to utilise AI to enhance the capacity for patient care. We Members of this profession must play an active discuss the other areas where this technology can role in the development, implementation, safe use show great potential in the patient care continuum and and validation of AI applications in medical imaging its implication to our practice. and radiation therapy. Radiographers need to be adequately equipped with the right skills to leverage AI utility as an optimisation tool has already gained these new technology-driven solutions. Doing so traction in imaging. In computed tomography (CT), allows practitioners to be informed of up-to-date deep learning image reconstruction (DLIR) and AI practices to their imaging and treatment practices algorithms allow the use of low radiation dose without and prepare for the change needed to adapt to the compromising final image quality. Newer CT models new technology. As gatekeepers in medical imaging, now carry a myriad of AI-based features designed to radiographers must understand the fallability of AI simplify and automate the CT workflows. An example technology, and that behind all that technology is a is auto-positioning of the patient to the isocentre of competent radiographer utilising it to deliver safe care the bore for different CT exams, reducing the risk of which benefits the patients. incorrect positioning and reduces the number of steps needed for patient positioning. Similar opportunities Digital competence of radiographers can be mapped for image optimisation and shorten scan time exists in as part of undergraduate education with elements MRI, which may be augmented with AI. The ability of AI of genomics, data analytics and AI embedded in the to improve precision leads to better turnaround time undergraduate curriculum. In this way, we can be and patient outcome. confident that radiographers will have the foundation and appreciation for such technological advancements. The use of AI-powered chatbots is benefits both patients and providers. Chatbots provides the While AI can take over mundane tasks and free up ability to connect with patients and is commonly radiographers, it can never replace the human touch. used to understand patient queries related to their It will be a matter of time until AI will be the new appointment, schedule appointment, sending operative, the new norm, in the delivery of care in reminders to patients and triaging a patient’s condition the imaging department. Knowing how to interact, based on their inputs. This ability automates roles adapt and utilise such technology will ensure the traditionally manned by customer support. AI chatbots sustainability of the profession. can also be extended to clinical staffs. For instance, during the COVID-19 pandemic, AI chatbots were Harris Abdul Razak introduced to National University Hospital (NUH) Singapore Society of Radiographers clinical staffs to retrieve vital information rapidly, such as COVID-19 protocols and hospital guidelines. High availability of complex data for clinical and non-clinical decision-making maximises the role of clinical staffs in improving patient pathways.
We have seen an explosion of AI tools being developed and validated to aid radiologists in confirming COVID-19 infections. Another emerging AI trend is on radiotherapy treatment planning which promises a more automated segmentation of tumours and sensitive organs. Existing methods for the same task takes days for a planner to design for each patient. AI technology will continue to evolve and be a disruptive force in medicine. The radiography profession must
42 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 Impact of AI on patient care in CT: the aspects of AI in CT that impact on patient care, Trinidad Article is a “Collaborative piece” from The Society of consideration should be given to some patient care Radiographers of Trinidad and Tobago concerns in CT. The scanner with which we work, & Tobago becomes in many instances, an extension of ourselves; SINCE Computed Tomography (CT) was introduced however, what is second nature to us can be very in 1973, the one constant in the industry has been intimidating to the patients we serve. As radiographers, that of change. CT evolution includes changes in CT we find ourselves having to act as a bridge between our generation, beam geometry, beam width, detector patients and this technology. Apart from technology, width, number of slices, detector technology, tube patient throughput, time constraints and complexity technology, rotation times and changes to image of tasks often put us at odds with patient centered reconstruction methods. Patients have been the main care and patient engagement.1,7 The provision of beneficiaries of these changes as the developments individualized care and health care encounters that are have been primarily aimed at shortening examination compassionate and patient focussed, where the patient time, lowering radiation dose, reducing contrast media, is helped to navigate the health care experience, with improving pathology detection and improving overall support, understanding and positive interactions, image quality and exam efficiency. These developments are all characteristics identified to have a profound have mainly been hardware related and the limits influence on the patient experience.1 While it is unlikely of hardware enhancements have been pursued and that patient volumes will diminish, any applications or pushed for almost 50 years. In today’s environment tools which are reliable and efficient and will liberate we see the impetus for further improvements being us from having to spend our time being equipment increasingly driven by software technology, enabled or technology focussed but instead allow us to be by advances in computer processing power. Artificial focussed on the patient, will lend themselves to Intelligence (AI) in particular, is the latest software enhancing patient care in CT. vehicle set to further revolutionize the delivery of CT, further streamlining efficiency, safety and AI systems are characteristically associated with some effectiveness. While CT technology has continued aspect of human intelligence. Examples include, but to evolve, so too has the notion of patient care. The are not limited to: learning, reasoning and problem expressions and definitions of “patient centered care”, solving. The term AI is used to refer to the use of “patient experience” and “patient engagement”1,7 computer systems to perform tasks that normally have emerged through the years as a testament to require human intelligence. Artificial intelligence can how quality of care received by our patients is defined. be split into two broad types: narrow AI and general Consequently, a review of the impact of AI on patient AI. The current AI applications commercially available care in CT should also mirror this scope. in CT can be considered as Narrow AI [8]. We see applications of this in everyday life: intelligent systems Before we delve into understanding AI and explore that have been taught or have learned to conduct
Figure 1: Deep Learning is a subset of Machine Learning (ML) while ML is a subtype of Artificial Intelligence. Artificial Intelligence allows machines to mimic human intelligence.
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 43 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW
specific tasks without being explicitly programmed networks, suited for different applications. For example, to perform that specific task. Machine intelligence recurrent neural networks are well suited to language of this kind is found in speech, language and object processing and speech recognition, while convolutional recognition of a virtual assistant on our mobile phones neural networks are more commonly used in image or on smart devices in our homes. These systems learn recognition.8 or are taught how to do specific tasks, hence the term narrow AI. General AI is very different, it is a Applications of AI in CT representation of flexible or adaptable form of human There are a number of current applications of AI in intelligence, capable of learning how to perform CT which ultimately impact on patient care. The main immensely different tasks.8 General AI does not exist areas of AI application commercially available are: and if or when it will become a reality is a topic of great Patient positioning, image reconstruction, detection & debate among AI experts. analysis.
Machine learning is a subtype of AI where a computer Patient Positioning system is fed large amounts of data; and it uses this AI is being used to assist radiographers with patient data to learn how to perform a specific task. The positioning accuracy in CT. An in room camera, system learns from data inputs to make decisions with mounted to the ceiling above the patient table, captures minimal human involvement; in other words the system patient’s dimensions while the patient is lying on the creates the rules or algorithms needed to achieve table.6 The table then moves into the correct position the task.8 Machine learning involves the creation of for scanning. The AI task is to position the patient in the neural networks. Networks of interconnected layers optimum position in the gantry based on the selected of algorithms, called neurons, that feed data into each protocol. Such a tool can shorten set-up times and other, and which can be trained to carry out specific improve image quality and radiation dose, owing to tasks. Training of these neural networks continue accurate iso-centric positioning.6 Images acquired by until the output from the neural network is consistent the camera are evaluated by AI algorithms.6 These with what is desired, at which point the network will algorithms have learned from a large amount of have ‘learned’ how to carry out that specific task.8 A clinical data comprising patients of difference sizes. 3D further subset of machine learning is deep learning. modelling of the patient’s position and orientation on Training occurs using massive amounts of data, the CT table is performed, then the system selects the therefore, artificial neural networks are expanded into ideal iso-center position.6 Manual fine tuning can also vast networks with an even greater number of layers. be performed if needed. These deep neural networks (DNN) contain multiple layers of mathematical equations with extensive Image Reconstruction connections trained and strengthened based on the A persistent dilemma in CT is that of lowering dose desired output.2,3 There are different types of neural while optimizing image quality. CT reconstruction
Figure 2. Deep Learning Neural Networks form multiple layers of mathematical equations with extensive connections trained based on the desired output. The ability to learn via a deep neural net gives Deep Learning algorithms the freedom to find the best way to achieve the required task. Conventional algorithms are limited by pre-programmed rules for performing a complex task, however, Deep Learning occurs when a neural network learns from its own intensive training process and develops its own reasoning based on this.3
44 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 techniques such as Filtered Back Projection (FBP) and the potential to improve patient outcomes through Iterative Reconstruction (IR) have their own challenges. early, accurate diagnosis of disease. FBP requires high radiation doses to produce high quality images. Meanwhile IR requires less dose than Apart from the applications described above, deep FBP but it possesses non-desirable image texture learning reconstruction is also being used in characteristics on the resultant image.2,3 Additionally, Spectral CT systems. One vendor uses rapid kV the application of more sophisticated IR techniques switching technology coupled with AI (Deep Learning such as Model Based Iterative Reconstruction comes at Spectral Reconstruction) to optimize energy a cost of reconstruction time which is disadvantageous separation and produce a low noise image. The to workflow efficiency and patient throughput.3 The Deep Learning Spectral reconstruction transforms introduction of AI in image reconstruction results multi-energy raw data sets (obtained by interleaved in improving image quality and preserving image acquisition of high kV views and low kV views as the texture in low dose images, compared with existing X-ray tube and detector rotate around the patient) into IR algorithms, without workflow compromise. Patient two full high and low energy separated sinograms. care benefits are related to lower radiation dose, faster Anatomical information contained in a high kV view patient throughput and improved image quality. and a low kV view at a particular location is common Deep Learning Reconstruction (DLR) is trained using to both views; spectral reconstruction uses this high quality target images. The exact target type is high spatial frequency information to its advantage, vendor dependent. Examples include, high quality resulting in energy separation for Spectral analysis FBP Target images2 and advanced MBIR Target with high resolution and low noise properties.4 The Images.3 The learning takes place at the factory and availability of spectral data permits spectral the system learns to turn high noise (low dose) input analysis opportunities for: the creation of data into low noise images. The specific goal is to monochromatic images (increasing iodine differentiate signal from noise and produce images contrast-to-noise ratio, decreasing metal or that are “sharp and clear”.3 This is a highly complex beam-hardening artifacts), the visualization and task and the machine learning involves an intensive quantification of iodine and other materials and the process using vast data inputs, leading to the characterisation and differentiation of tissue. This development of a DLR convolutional neural network data provides Radiologists with greater access to (CNN).3 Once trained the DLR operates in the raw and relevant diagnostic information leading potentially image domains to efficiently reconstruct images.3 to more definitive and confident diagnoses. Spectral Scanners with DLR capability usually have three levels imaging also provides the opportunity for increasing of application which allow images to be produced with patient safety through reduced contrast usage and the three levels of noise reduction.2,3 These settings can be omission of unenhanced CT scans.5 Deep Learning built into the protocol and saved based on Radiologist Spectral provides several opportunities for patient preference. This makes DLR easy to apply, not requiring care enhancement in CT through improved safety and additional technologist input. Any application that increased diagnostic capability. can be pre-set or not requiring operator intervention, promotes the opportunity for the radiographer to focus Detection and Analysis on the patient. AI is being applied to image interpretation solutions. AI in this area is aimed at workload reduction and Owing to the DLR capability, CT imaging using Ultra efficiency improvements. In some instances AI helps High Resolution Scanning (UHR), with 0.25mm x 160 with basic, repetitive tasks, in other instances it detector elements and 1024 image reconstruction helps with more complex interpretative tasks or matrix, is now possible with dose neutral imaging. where reporting is time sensitive or critical. AI DLR learns to preserve edges and maintain image solutions currently available have a number of detail and spatial resolution while improving noise commonalities: Firstly, Classification, where data and low contrast characteristics, without the need is automatically prepared and categorized based on to increase radiation dose or impede workflow. exam type then auto selection of the appropriate AI Scanners also incorporate DLR into auto mA application follows. Secondly, Analysis and Results, selection which further individualizes the patient’s where the AI application performs automatic analysis, dose based on the dose reduction abilities of deep producing structured results, this can involve learning reconstruction.3 Access to dose neutral UHR identifying and quantifying relevant anatomy and CT scanning provides several advantages including pathologies and placing findings in diagnostic context.9 the ability to: evaluate small arterial vasculature in Flagging of potentially urgent conditions can also peripheral CT angiography10; evaluate pancreatic occur. The goal of these platforms is generally to tumors with the ability to assess local tumor invasion; permit faster, more accurate diagnosis and reduce detect solid lesions as well as subtle ground glass turn-around time for reports to improve patient opacities in the lung11,12; quantification of coronary outcomes. Some clinical applications include: stenosis13 and airway measurement14. These all have
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 45 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW
Chest AI: images for flagging and communication of suspected AI Applications are aimed at detecting lesions, positive findings of linear radiolucencies in the bones determining risk (in the case of Calcium Scoring), of the cervical spine identifying patterns compatible measuring vessel diameter, detecting emphysema in with fractures.16 Also available is quantitative and the lungs, detecting pulmonary emboli, performing qualitative analysis in the evaluation and assessment of segmentation and analysis. AI platforms automatically musculoskeletal disease.16 generate a standardized quantitative report of the data.9 Using ECG-gated CT scans, the degree of calcification Oncology AI: in a patient’s coronary arteries can be calculated. Application for viewing, manipulation, 3D- visualization Narrowing of the coronary arteries can lead to adverse and comparison of medical images from multiple cardiovascular events. Radiologists can better diagnose imaging modalities and/or multiple time-points.16 the severity of defects, resulting in improved patient CT image software is also available for analysis outcomes.15 and visualization of Liver CT data. It is designed for assessing liver morphology with automated tools for Additionally, radiologists generally center their CT liver segmentation and measurement.16 interpretations based on the primary clinical indication. The lung AI platform is trained to perform a systematic Conclusion and equal examination of all areas of the chest, AI is poised to continue having a significant impact on potentially finding lesions in regions that the reader CT. AI has already been established across three major may not have made a priority.9 areas of CT: patient positioning, image reconstruction and image analysis. These applications all have a In CT studies containing the lung or a portion of direct impact on patient care. They result in greater the lung, such as, scans of the chest, abdomen or examination efficiency; improvement in patient safety; cervical spine, AI for the detection and prioritization of reduction in radiation dose; enhancement in image incidental COVID-19 findings is available.17 Prioritization quality; and faster, more accurate diagnosis. AI enables of incidental findings of ground glass opacification, a increased automation of various activities considered non-specific imaging finding associated with COVID-19 tedious, complex, time sensitive or high-stake; allowing infection, may help in the management of adverse greater focus to be on the patient. As AI development effects of COVID-19.17 A patient undergoing oncology continues, newer, more advanced applications will screening or trauma imaging, for example, but who provide further efficacy and efficiency. The onus will be does not exhibit any COVID-associated respiratory on Radiographers and other Health Care Professionals symptoms, will benefit from incidental COVID-19 to embrace these changes and ensure that they use findings as this information can promote further these opportunities, to provide the quality of care our patient evaluation.17 Prompt identification of incidental patients deserve. COVID-19 findings will permit sooner treatment or isolation and will help limit the spread of the disease. References 1. Bolderston, A. (2016). Patient Experience in Medical Brain & Stroke AI: Imaging and Radiation Therapy, Journal of Medical AI performance features in this area include: Automatic Imaging and Radiation Sciences. 47, 356-361 labelling, visualization and volumetric quantification of 2. Hsieh, J., Liu, E., Nett, B., et al. (2019). A new era segmentable structures from CT images of the brain of image reconstruction: TrueFidelity™. Technical to provide automated registration and reformatting of White Paper on deep learning image reconstruction. data.16 Other AI platforms, review images immediately GE Healthcare. after the patient is scanned and informs radiologists 3. Boedeker, K. (2019). AiCE Deep Learning of potentially dangerous cases such as intracranial Reconstruction: Bringing the power of Ultra-High hemorrhages, as this can be life-threatening, therefore Resolution, CT to routine imaging. White Paper, the prioritization function is important. This deep Canon Medical Systems Corporation. learning application helps to reduce turnaround time 4. Boedeker, K., Hayes, M., Zhou, J., et al. (2019). Deep by analyzing images and increasing radiologists’ Learning Spectral CT – Faster, easier and more confidence in decision-making.15 AI-based automation intelligent. White Paper, Canon Medical Systems platforms for stroke analysis are zero-click solutions Corporation. that use deep learning technology to streamline the 5. Woo Goo, H., Mo Goo, J. (2017) Dual-Energy CT: workflow for fast results. Deep learning technologies New Horizon in Medical Imaging. Korean Journal of process and deliver images for accurate triage, Radiology. Jul-Aug; 18(4): 555–569 prioritization and decision making. 6. How Artificial Intelligence Can Improve CT Scans. Available at: https://new.siemens.com/global/en/ MSK AI: company/stories/research-technologies/artificial- AI Software available include triage and notification intelligence/artificial-intelligence-imaging- software used in the analysis of cervical spine CT techniques.html Accessed June 28, 2020
46 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 7. The Eight Principles of Patient-Centered Care. Available at: www.oneviewhealthcare.com/ the-eight-principles-of-patient-centered-care/ Accessed June 28, 2020 8. What is Artificial Intelligence (AI)? Available at: www.zdnet.com/article/what-is-ai-everything- you-need-to-know-about-artificial-intelligence/ Accessed June 26, 2020 9. SOMATOM X.cite with myExam Companion. Intelligent Imaging. Excellence Empowered. Available at: https://static.healthcare.siemens.com/ siemens_hwem-hwem_ssxa_websites-context- root/wcm/idc/groups/public/@global/@imaging/@ ct/documents/download/mda5/mzy0/~edisp/ siemens-healthineers-ct-single-source-somatom- xcite-06826346.pdf Accessed June 26, 2020 10. Tanaka, R., Yoshioka, K., Takagi, H., et al. (2018) Novel developments in non-invasive imaging of peripheral arterial disease with CT: experience with state-of-the-art, ultra-high-resolution CT and subtraction imaging. Clinical Radiology. Apr 4. doi: 10.1016/j.crad.2018.03.002. 11. Hata, A., Yanagawa, M., Honda, O., et al. (2017). Effect of Matrix Size on the Image Quality of Ultra- high-resolution CT of the Lung: Comparison of 512 x512, 1024 x1024 and 2048 x 2048. AcadRadiol, doi. org/10.1016/j.acra.2017.11.017 12. Yanagawa, M., Hata, A., Honda, O., et al. (2018). Subjective and Objective Spatial Resolution Comparisons between Ultra-High Resolution CT and Conventional Area Detector CT in Phantoms and Cadaveric Human Lungs. European Radiology. 13. Takagi, H., Tanaka, R., Nagata, K., et al. (2018). Diagnostic Performance of Coronary Angiography with Ultra-High-Resolution CT: Comparison with Invasive Coronary Angiography. European Journal of Radiology. /doi.org/10.1016/j.ejrad.2018.01.030 14. Tanabe, N., Oguma, T., Sato, S. et al. (2018). Quantitative Measurement of Airway Dimensions using Ultra-High Resolution Computed. Respiratory Investigation. 56; 489–496 15. 5 FDA Approved Uses of AI in Healthcare www.docwirenews.com/docwire-pick/future- of-medicine-picks/fda-approved-uses-of-ai-in- healthcare/ Accessed July 03, 2020 16. FDA Cleared AI Algorithms www.acrdsi.org/DSI- Services/FDA-Cleared-AI-Algorithms Accessed July 03, 2020 17. FDA Approves Use of Aidoc’s AI Algorithms for Incidental CT Findings Associated with COVID-19 www.itnonline.com/content/fda-approves-use- aidoc%E2%80%99s-ai-algorithms-incidental-ct- findings-associated-covid-19 Accessed July 03, 2020
Mrs Suzette Thomas-Rodriguez and Ms Aneesa Ali
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 47 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW
Artificial Intelligence and its application in the professionals, an open access artificial intelligence Tunisia radiology departments in Tunisia: tool to help instantly diagnose the disease from simple A route to be defined X-Rays of the lungs thus accelerating screening in regions without medical centers. This online platform THE evolution of technology, the aging of populations has enabled our Tunisian hospitals to detect COVID-19 and the lack of financial and human resources in health in record time with an accuracy rate of up to 92% and at represent a great challenge for the health systems low cost (Figure 2). which force it to optimize their resources. Using new technologies, such as Artificial Intelligence (AI), is A Tunisian start-up also succeeded in carrying out beneficial for patients and healthcare professionals. a functional characterization of the SARS-CoV2 By using large amounts of data generated by the genes using an AI that it developed in Tunisia and different medical imaging methods, AI will allow us even managed to identify candidate molecules last to increase our capacity for analysis and decision- February (Figure 3). Tunisian radiology technologists making, to improve the way we diagnose, detect, treat, are, therefore, among the professionals who directly predict a disease and support a patient.1 In medical participate in the success of this technology by the imaging, the radiology technologist will participate quality of the radiological images they produce daily. in the management of AI projects such as improving the workflow, producing high quality images, image AI committees and roadmap segmentation, improving image interpretation, Committees have been formed to study the impact automation of image recording, radiomic analysis of AI on the health professions including that of the (Figure1). radiology technologist. In conjunction with recent union movements to improve the situation of paramedical AI and medical imagery in Tunisia professionals, other working groups made up of In medical imaging, AI finds a place already ready former radiology professionals have been brought with the use of digitization. It is now possible to have a together to provide ideas on the role of the technologist diagnosis, which can be, sometimes more precise than in the application of AI projects in medical imaging a report from a radiologist. But, with the application of departments. Since this technology has a direct interest this technology in our radiology departments, how can in the field of radiology, the place of the medical we take advantage of AI to improve patient care? Will imaging technologist is important in the planning, our technical work also be transformed? What place execution and quality control of data. The technologist will radiology technologists have in this area? Questions will be responsible for the organization, classification that several health committees in Tunisia are trying of databases and segmentation of DICOM images to answer by developing a roadmap whose purpose (Figure 4). is to specify the new responsibilities of radiology technologist, like other health professionals. After some time of reflection and research, the In an African country like Tunisia, the use of AI was not committees concluded that group work is essential new. During the COVID-19 crisis, for example, Tunisian in the field of AI. They suggested several actions to engineers developed, in collaboration with radiology change the work of health professionals and at the
Figure 1: AI areas of impact for medical imaging practice.
48 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 Figure 2: Diagnosis of COVID-19 by AI using radiographs of the lungs performed by radiology technicians.
same time improve patient care. Here are some of the when using AI. main recommendations: • Sensitize technologists on ethical and legislative • Encourage research and strengthen the partnership rules. between students, researchers and industry. • Privilege informed communication with patients • Strengthen quality control in medical imaging since to avoid ambiguity in the use of AI and have their DICOM images are the basis of several AI projects in confidence. health. • Establish collaboration between foreign health and • Encourage the culture of innovation among radiology research bodies to exchange ideas and seek avenues technologists by setting prices and grants. for partnership. • Finance AI projects that add value to the quality of radiological examinations. The radiology technologist and the AI • Add AI in the academic training of the radiology In Tunisia, several discussions are underway for a technologist. complete reorganization of the category of paramedics • Add the theme AI in the permanent scientific training including radiology technologists. The goal is to allow during the national congresses. technologists to benefit from the implementation • Study the possibility of establishing a university path of new technologies, such as AI. New horizons applied to health in the field of AI. and additional tasks will therefore be added to the • Introduce radiological protocols based on the technologist’s profession. automation of technical tasks. • Collaborate with radiologists to multiply AI projects After committee work, a declaration from the Tunisian in radiology departments. Society of Radiology and the Tunisian Association of • Continue to strengthen the responsibility for Radiology Technologists will not be long in coming to radiation protection for the radiology technician raise awareness on important topics related to the use
Figure 3: An artificial intelligence tool to help screen for COVID-19.
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 49 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW
Figure 4: Technologist’s place in an AI project team.
of AI as ethics, technologist responsibility, radiation the future. Journal of Medical Radiation Sciences. protection, communication with patients. 2019. Volume: 66, Issue: 4, Pages: 292-295 https://onlinelibrary.wiley.com/doi/full/10.1002/ Particular attention to an important task for the jmrs.369#jmrs369-fig-0001 technologist: the development of radiological protocols 2. Covid-19 diagnosis by AI. Innov Challenge. is at the heart of the changes caused by AI. Thus, www.innovchallenge.com/ it is essential that clear protocols are developed to 3. Le spécialiste. Tunisie: un outil d’intelligence support the implementation of new systems and, where artificielle pour aider au dépistage. 17 avril 2020. possible, quality standards are developed jointly by www.lespecialiste.be/fr/actualites/tunisie-un-outil- professionals in imaging and radiotherapy services to d-intelligence-artificielle-pour-aider-au-depistage. support the implementation consistent implementation html of proven technologies to high standards.3 Mohamed Khelifi, References Medical Imaging Coordonnator 1. Gravel. P. L’intelligence artificielle à la rescousse Council member ISRRT, Tunisia du système de santé. Le Devoir 2019.www. LaSalle Hospital, Montreal, Canada ledevoir.com/societe/science/556361/l- intelligence-artificielle-a-la-rescousse-du- systeme-de sante#:~:text=L’avantage%20 de%20l’IA,am%C3%A9lioreront%20le%20 syst%C3%A8me%20de%20sant%C3%A9.%20 %C2%BB) 2. International Society of Radiographers and Radiological Technologists, European Federation of Radiographer Societies. Artificial Intelligence and the Radiographer/Radiological Technologist Profession: A joint statement. Radiography. 2020. Volume 26, Issue 2, P93-95, www.radiographyonline. com/article/S1078-8174(20)30037-7/fulltext)
Figures 1. Lewis. S. Gandomkar. Z. Brennan. P. Artificial Intelligence in medical imaging practice: looking to
50 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 Elevating patient care with Artificial Intelligence: differences and patient preferences before, during United Radiographers are essential in elevating patient and after the examination. care with Artificial Intelligence Kingdom 2. Customised and improved examination experience, The UK perspective including optimal positioning and data acquisition Artificial Intelligence (AI), an ever growing field of techniques based on patient biometry, gender, age, computer science, working to construct and validate area of interest and condition, to improve the quality smart machines and smart algorithms that are of the imaging examinations, minimise repeats and capable to perform tasks typically requiring human technical errors from (motion) artefacts and make intelligence1,2, is already here. AI and all its different examinations better tolerated by patients (e.g. AI sub-fields such as machine learning and deep learning can reduce the need of breath-holds for MRI scans, have indeed already permeated clinical practice in correct retrospectively for motion artefacts or even Medicine, as well as Science and Technology3-5. reduce the duration of radiotherapy sessions, if planned using precision medicine11). Radiography, both diagnostic and therapeutic, has always been at the forefront of technological 3. Consideration of fully integrated radiomics and innovation and the workforce has built resilience and genomics, which would offer a complete picture a unique level of adaptation to and engagement with for the patient but also comparison of findings with technological advancements6, by ensuring research and patients of a similar profile. These could inform education follow suit to support the changing practice. optimal patient management and facilitate improved patient outcomes, allow for more comprehensive Radiography, both diagnostic and therapeutic, being holistic, person-centred examinations, addressing one of the most technology-enabled professions, is the person/patient and not only their condition in experiencing a groundbreaking transformation in isolation. It would also allow the patient to make clinical practice because of AI: this includes workflows, truly informed choices in relation to acceptance of patient data management, archiving and storage, imaging and treatment,. patient positioning and image acquisition and planning, motion correction, image segmentation and post- 4. Optimisation of workflows, to minimise delays processing but also image interpretation7-9. Smart in referrals, facilitate quicker turnaround times planning tools in current clinical use, for instance, of reports and more robust results in image include automated rule implementation and reasoning, interpretation. This could also extend to include modelling of prior knowledge in clinical practice, and same day referral onto further imaging or prompt multi-criteria optimisation. There are unique features commencement of treatment pathways. Artificial involved in AI that make it perhaps one of the most Intelligence systems could simplify many steps revolutionary technology advancements our profession of the complex workflow of radiotherapy such as will probably, with many claiming this is indeed segmentation, planning or delivery12. disruptive innovation, which is trained to learn and organically grow and adapt exactly like the human brain 5. Increased patient safety profiles, with more and intelligence does10. standardised clinical governance, quality assurance and quality control checks, universal safety Like every technological innovation, it is important to indicators, customised to each examination and ensure it is used as a tool to augment and enhance individual clinical case. practice, patient care and safety and quality of imaging, the very purpose of the profession of radiography. Those who dream of faster workflows and higher Changes in radiography practice brought over by throughputs of patients undergoing medical imaging AI remain currently – mostly- isolated in specialist examinations and radiotherapy treatment must not or research centres, but they increasingly start look only to AI for this, as staff burnout and reduced job impacting clinical work. These are translated into more satisfaction will lurk around the corner to negate any standardisation and increased workflow efficiency, but potential benefits. Instead, faster throughput of patients there is now more than ever scope to ensure that the and more efficient workflows can only be assured savings in time can be used to improve patient care in with concurrent significant investment in advanced many different ways. These can include some of the education and training into the new technologies of following: medical imaging staff (for radiographers, radiologists, 1. Better communication both between patient medical physicists), proportionate staff recruitment and and clinical teams but also between the clinical the procurement of high-end software and hardware; teams; this may include seamless interdisciplinary otherwise the benefits of AI can quickly evaporate into communication of patient’s condition and needs, thin air. communication with patient, family and carers and customisation of the approach to respect cultural There are many challenges standing in the way of an
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 51 THEMATIC APPROACH – PRACTICING RADIOGRAPHER’S VIEW
effective implementation of AI in the UK and other There is currently a small but increasing number of countries13-15. postgraduate and CPD educational initiatives on AI offered by UK Universities, but these need to multiply The changes in clinical practice of radiography due to and allow more flexibility for the workforce that is AI can already be felt, as described earlier here; currently practicing but also for current students and however this progress is barely matched by future/imminent graduates who will be faced with a proportionate changes in education and training but much different reality than their predecessors. also in relevant radiography-led research projects, with few bright exceptions in the UK academia and There is a small but increasing number of UK beyond. Thankfully, but slowly, this trend seems academic conferences, seminars and online resources to change and attitudes to shift. It is only through focused on AI and aimed at raising awareness of practice-led research and research-informed teaching radiographers on AI topics, and allowing discussions that AI can be effectively implemented and the fruit of of contemporary issues in this field. Many of these what is now only AI promises can truly materialise for resources are also provided by Health Education the benefit of the patients and the clinical teams that England and the Department of Health to support the serve them. training of the workforce, inline with the Topol review recommendations16. More collaboration between A better-educated - on the basics of AI and patient- clinical practice and academia will ensure these centred care- radiography workforce can act as educational provisions will help align theory with the gatekeeper and identify a faulty or not-properly practice and make training relevant and meaningful to validated algorithm and highlight when results might the workforce. not make clinical sense for the patient or examination in question. An up-to-date radiography practitioner Some small isolated local (often single-site) research can perform regular quality control and quality projects on AI led by UK radiography academics are assurance of AI software and hardware and interpret in place but we need a robust framework, which will findings, offering solutions to streamline workflows. A allow large, multidisciplinary teams to form and work radiographer who is adequately educated to understand together on big project grants and academic and AI can also help the clinical validation of this algorithm research radiographers to have enough notice and in practice, as we are after all the workforce that support to collaborate towards common targets and spends more time with the patients in medical imaging priorities for the profession, as identified by the Society examinations. and College of radiographers research priorities for the profession17. A mapping of current AI projects Research stemming from within Radiography will also in the UK by a central agency would be essential to support the integration of high-quality, truly validated, ensure there is coordination and we avoid unnecessary AI-supported medical imaging and treatment with duplication of effort; whose role this might be, remains patient-centred care. It will also ensure it brings the yet to be answered. patient voice into the equation to help inform the design of medical imaging equipment and processes. In this Lobbying and inclusion of radiographers in the AI role the radiographers will act not as ambassadors planning and decision-making committees will be or advocates but will ensure the patients participate vital to ensure a proportionate representation of our as equal research partners into the design of the profession in a future with AI. This agenda is currently equipment and the mapping of the processes that will being supported by our professional body, the Society be used for their imaging examination. and College of Radiographers, who have fostered initiatives to support a review of the current education Participation of radiographers in national and and careers framework (ECF) with the inclusion of international consortia and decision-making many experts from different modalities in the field and committees for AI is not currently really representative with AI as a key theme, among other priorities. of the workforce size or scope or indeed of its unique role in managing and facilitating this transition to a Artificial intelligence is just a tool and subject to future with AI. Unfortunately this means that there is a regulated and controlled use it can really revolutionise real danger that we could miss out on the opportunity not only medical imaging and radiography but also all to inform the future of our practice, education and fields of healthcare, for the benefits of the patients. research in this field if we are not part of the decision Like any tool, its use has to be validated, quality- making process. controlled, subject to the same scrutiny and ethics, as previous healthcare technologies for it to be deemed Some progress to address some of these challenges safe for use and clinically meaningful. Patient input in the UK is currently in place, which is reassuring, but will be vital in the outputs of AI software and hardware this momentum has to intensify. to ensure it serves patient needs and allows for customisation of radiography service. Education and
52 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 research in radiography will need to accelerate to radiographers-research-priorities-radiographic- ensure the workforce is equipped with what they need profession (accessed 24th July 2020) to make the most out of this amazing new opportunity and to minimise any risks that may arise now and in This article on behalf of the SCOR was written by the future. Dr Christina Malamateniou Director of Postgraduate and Doctorate Programme Acknowledgements in Radiography Mr Richard Evans, OBE, Dr Rachel Harris, Dr Tracy City, University of London, O’Regan, Dr Spencer Goodman and Charlotte [email protected] Breadmore from the Society and College of Radiographers for their critical comments on this document and for their support and leadership in pushing the AI agenda forward for Radiography.
Many thanks to everyone in the Department of Radiography at City, University of London, students and staff as well as our many clinical partners for insightful discussions and for sharing their queries, anxieties and ideas on AI and technological innovations so generously with us over the past few years.
References 1. www.britannica.com/technology/artificial- intelligence (accessed July 1st 2020) 2. www.aaai.org/Organization/presidential-panel.php (accessed July 30th 2020) 3. https://link.springer.com/article/10.1007/s00146- 016-0685-0 (accessed July 1st 2020) 4. www.tandfonline.com/doi/abs/10.1080/13645706.20 19.1575882 (accessed July 7th 2020) 5. www.ncbi.nlm.nih.gov/pmc/articles/PMC6199205/ (accessed July 20th 2020) 6. www.sor.org/ezines/scortalk/issue-76/ai-radiology- key-feature-topol-review (accessed July 20th 2020) 7. https://onlinelibrary.wiley.com/doi/full/10.1002/ jmrs.369 (accessed July 15th 2020) 8. www.birpublications.org/doi/abs/10.1259/ bjr.20190840?journalCode=bjr (accessed July 15th 2020) 9. www.ncbi.nlm.nih.gov/pmc/articles/PMC6732844/ (accessed July 30th 2020) 10. https://link.springer.com/ chapter/10.1007/978-3-030-35975-1_1 (accessed July 20th 2020) 11. https://link.springer.com/ chapter/10.1007/978-3-030-27994-3_5 (accessed July 20th 2020) 12. https://link.springer.com/article/10.1007/s12032- 020-01374-w 13. www.isrrt.org/artificial-intelligence (accessed July 15th 2020) 14. https://bmcmedicine.biomedcentral.com/ articles/10.1186/s12916-019-1426-2 (accessed July 17th) 15. www.jacr.org/article/S1546-1440(19)30699-4/pdf (accessed July 20th 2020) 16. www.hee.nhs.uk/our-work/topol-review (accessed July 7th 2020) 17. www.sor.org/learning/document-library/college-
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 53 CONTRIBUTING STAKEHOLDERS VIEWS
Artificial Intelligence – A challenge and an opportunity for the radiography and radiology professions Adrian Brady Adrian Brady is a Consultant Radiologist in The Mercy University By Adrian Brady Hospital, in Cork, Ireland, and a Clinical Senior Lecturer at University College Cork. He graduated in medicine DEPENDING on your viewpoint, (fundamentally digital) data. It’s not a from University College Dublin 42 this apparently simple number huge step for us to expand our horizons in 1984. After pursuing specialty may signify many different things: the to encompass the exciting developments training in internal medicine, he atomic number of molybdenum, the sum in machine learning, data mining and AI began training as a radiologist in of the first six positive even numbers, application which are emerging on an Dublin in 1987. In 1991, he moved the number with which God creates the almost-daily basis. to Canada for further training in universe in Kabbalistic tradition. Perhaps gastrointestinal and abdominal it is best known in modern culture as the Peruse the contents pages of any current imaging and intervention, at “Answer to the Ultimate Question of Life, significant radiology or radiography journal; McMaster University Medical the Universe and Everything”, calculated one of the most-represented categories of Centre, in Hamilton, Ontario. over 7.5 million years by the enormous papers is informatics and AI. This is not a He worked as a staff radiologist supercomputer named Deep Thought (in temporary flash in the pan. We should not in the Wellesley Hospital, Toronto, Douglas Adams’ ‘The Hitchhiker’s Guide to assume that after the first flush of papers, and Assistant Professor at the the Galaxy’1). Sadly, no-one knows what the AI will diminish in importance, and things University of Toronto from 1992 to question is. will go back to where they were before. AI 1995, returning to Ireland in 1995 is already changing what we do and how we to take up his current position, Adams (and his readers) had much fun do it, and this change will accelerate. Many where he works as a general and with this answer/question difficulty, of the time-consuming tasks we perform interventional radiologist. but underpinning the entire theme is a can be done more rapidly and with greater Dr Brady has served in many truth: much of our lives is governed by sensitivity and efficiency by AI tools. AI is leadership positions within Ireland mathematics. As Artificial Intelligence already capable of contributing to our work and Europe, serving as Dean of develops, and increases its involvement in in a multitude of ways. Lesion detection the Faculty of Radiologists, RCSI healthcare, this digitization of who we are and segmentation can be done (at least in (the equivalent of President of and what we do will accelerate. some radiologic studies) by algorithms. the National society and head of Imaging protocol selection can be done national radiology training) from For professionals such as radiographers in real time by AI (e.g. findings detected 2010 to 2012. and radiologists, this may not be as large automatically on early sequences of an MR He was chair of the ESR Quality , a transition as for some other healthcare study can automatically lead to parameter Safety and Standarts Comitee from workers. Our working environments are or sequence adaptation later in the study 2017 to 2020 and is now ESR 2nd already physics-rich, and heavily dependent to optimise the imaging). Workflow can be Vice President. on our technical knowledge and expertise. changed and improved by AI (e.g. identifying We work on a daily basis with human- patients more likely not to keep an digital interfaces and large amounts of appointment, and automatically reminding digital data. Our specialties are thus ideally them). positioned to reap the benefits of current rapid developments in AI application to Some of these innovations are medicine, and we have a great opportunity straightforward, and unequivocally to be in the vanguard of informatics and AI welcome. We may be more ambivalent developments in medicine. We are already about others. Many available software in the business of generating, manipulating packages are better than we humans are at and interpreting large amounts of identifying and highlighting lung nodules on
54 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 CT. However, that doesn’t mean these consent is always ensured for any use of European Society of Radiology white AI tools are better than we are at their personal data. AI opens up a plethora paper: What the radiologist should know performing the higher-level synthetic of potential new uses for patient data, for about artificial intelligence.Insights tasks of interpreting the meaning of clinical work, software development, data into Imaging (2019);10:44. DOI: 10.1186/ these nodules and formulating rational mining and radiomics and commercial s13244-019-0738-2 differential diagnoses. The value of AI in development. It’s incumbent on those 4. Brady AP, Neri E. Artificial intelligence the future is likely to be in assisting us professionals such as radiographers, in radiology – ethical considerations. by drawing our attention to findings, and radiological technologists and radiologists Diagnostics 2020;10(4):231. DOI: providing us with information we might not who are central to the creation, 10.3390/diagnostics10040231 have perceived; there will still be room for, manipulation and interpretation of that 5. European Society and European and a definite need for human involvement data to ensure that it is only used for the Federation of Radiographer Societies. in interpretive tasks. benefit (direct or indirect) of those whose Patient safety in medical imaging: a joint information we are privileged to use, and paper of the ESR and EFRS. Insights AI isn’t magic, it’s maths. While much of always with their consent6. into Imaging (2019)10:45. DOI: 10.1186/ the natural world, from fundamental forces s13244-019-0721-y, and Radiography to cosmology, is governed by mathematical In terms of patient safety, the most- (2019);25(2):e26-e38. relationships and rules, few of us would minimal directive is the “requirement 6. International Society of Radiographers be willing to cede decision-making about that all possible efforts should be made and Radiological Technologists and The the important things in our lives to such to ensure patients are no worse off after European Federation of Radiographer rules. Although it may impress with its their interaction with radiographers and Societies. Artificial Intelligence and the speed and efficiency, AI is not an empathic radiologists than before”7. Naturally, we radiographer/radiological technologist organism, sensitive to the emotional needs always seek to go far beyond this basic profession: a joint statement of the of others, or conscious in any meaningful limit. AI offers wonderful opportunities for International Society of Radiographers way. For those attributes, we need humans. patients, and for professionals to improve and Radiological Technologists and the We must not lose sight of the basic fact the quality of the services we offer. It also European federation of Radiographer that AI algorithms are mathematical carries risks, which can best be mitigated Societies. Radiography (26);2020:93- functions, designed (or self-designing) by keeping knowledgeable professional 95. https://doi.org/10.1016/j. to fulfil pre-determined functions or humans in charge6. Radiographers, radi.2020.03.007 achieve programmed outcomes. It’s easy radiological technologists and radiologists 7. McNulty JP, Brady AP. Editorial: to imagine an AI programmed to eliminate are ideal intermediaries in this new world, Patient safety: at the centre of all cancer in humans. The most-direct, able to harness our clinical knowledge and we do. Radiography (2019);25(2):99- simplest and fastest way to achieve that humanity and our technological expertise 100. https://doi.org/10.1016/j. goal is to eliminate the human race, but to benefit and protect our patients. So, to radi.2019.03.003 that’s probably not what the algorithm’s return to that fountainhead of wisdom, designers wanted or what the population Douglas Adams: “Don’t Panic”1. on which the AI is deployed might welcome. This article was written by Adrian Brady, on Just because an outcome is true to the It’s a pleasure and privilege to be able to behalf of the ESR. mathematics of the AI doesn’t mean it’s make this contribution on behalf of the ESR desirable, and steadying human hands will to the ISRRT annual special edition on the always be needed, if AI is to prove a boon occasion of World Radiographer’s Day, and to humanity. to reflect on and emphasise the vital role our professions jointly fulfil in patient care, So, AI is versatile, clever and capable of safety and welfare. things we humans can’t do. But it’s not a replacement for us, and humans must References: remain in control of AI development 1. Adams D. The Hitchhiker’s Guide to the and use. There are fundamental ethical Galaxy. 1979, Pan Books, London. principles that should underpin AI, some 2. Geis R, Brady AP, Wu CC, Spencer J, of which have been elaborated upon in the Ranschaert E, Jaremko JL, Langer SG, multisociety radiology society statement Borondy Kitts A, Birch J, Shields WF, on the ethics of AI in radiology, published van den Hoven can Genderen R, Kotter in 20192, and in other related papers3,4. E, Wawira Gichoya J, Cook TS, Morgan Also in 2019, the EFRS and ESR jointly MB, Tang A, Safdar NM, Kohli M. Ethics published a statement on patient safety of AI in Radiology: Summary of the in medical imaging5. Among the aspects Joint European and North American of safety we considered in that paper was Multisociety Statement. Insights into data security, the need to ensure that Imaging 2019;10:101. https://doi. imaging data access is only provided to org/10.1186/s13244-019-0785-8. appropriate individuals, and that patient 3. European Society of Radiology.
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 55 CONTRIBUTING STAKEHOLDERS VIEWS
Medical Imaging and radiotherapy radiographers have a responsibility to ensure Artificial Intelligence is Nick Woznitza Consultant Radiographer, used to elevate patient care: The Homerton University Hospital, London, UK views of the European Federation of School of Allied and Public Health Professions, Canterbury Christ Radiographer Societies Church University, UK Co-Chair EFRS-ISRRT Artificial Intelligence Working Group By Nick Woznitza, Spencer Goodman, Jonathan McNulty
RADIOGRAPHERS have always been the be able to critically evaluate and appraise patient facing element of medical imaging all interventions in medical imaging and radiotherapy, at the intersection of and radiotherapy, to ensure that patient technology and patients. As a profession, outcomes are improved where AI is radiographers have always evolved and introduced into clinical practice. adapted to emerging technologies; this has not changed during the artificial Are radiographers involved in the intelligence (AI) “revolution” but perhaps design and implementation of AI in the scale and speed of change has clinical practice? Material arising from increased. medical imaging industry partners is Spencer Goodman often solely focused on the impact on Professional Officer for As healthcare professionals it is essential radiologist workflows and experience. Radiotherapy, that radiographer practice is evidence- Without radiographers obtaining high Society and College of based. Recent high-profile retractions from quality diagnostic examinations there Radiographers, UK New England Journal of Medicine (Mehra is, however, no data for radiologists et al., 2020) and Lancet (Mehra et al. 2020) or algorithms to interpret! It is just related to the COVID-19 pandemic reinforce essential that radiographers contribute the requirement for thorough and robust to such developments. The recent joint evaluation of healthcare, with transparency. statement of the International Society AI, even in a decision support role, can have of Radiographers and Radiological an impact on patient outcomes and needs Technologists (ISRRT) and the European to be evaluated with similar diligence Federation of Radiographer Societies to medication. “Algorithms are the new (EFRS) on Artificial Intelligence and the drugs” and can have a profound impact Radiographer Profession, concluded that: on care decisions (Harvey, 2017). A recent “... of critical importance that systematic review highlighted the poor- radiographers, as medical imaging quality published research underpinning and radiotherapy experts, must Jonathan McNulty many AI healthcare interventions play an active role in the planning, (Nagendran et al., 2020) and we need to development, implementation, use President, European Federation of assure ourselves and our patients that and validation of AI applications Radiographer Societies we are providing safe and effective care. in medical imaging and radiation Associate Dean, That is not to say that we should dismiss therapy, reinforcing the need for the School of Medicine, all new interventions, but rather introduce technology to be targeted to the most University College Dublin, Ireland changes to practices in a controlled and pressing clinical problems.” systematic way. Radiographers need to (SRRT & EFRS, 2020)
56 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 Various algorithms are being developed model of patients at the heart of planning responsible for delivering the radiation and marketed that recreate “high quality” and service delivery will ensure trust and dose to the patient. As radiographers, are CT and MRI images with reduced radiation engagement for subsequent data collection we aware of how these “suggestions” are dose or scan time using algorithms to and improvement of any models. Figure developed? The 2019 American Society improve the signal to noise ratio. But, as 1 highlights the key stakeholders in any of Radiologic Technologists (ASRT) radiographers, are we assured that the ‘partnerships’ related to the development survey found that radiographers had information that we are generating is a and implementation of any clinical service mixed familiarity with AI features on their true representation of the patient we are delivery interventions. equipment; however, a majority were examining and not an “average” patient? confident it functions correctly and felt How do we know a rare or significant Importantly for AI, PPI or patient / results were trustworthy (ASRT, 2019). abnormality has not been replaced on the practitioner partnership for both medical Radiographers need to be aware of, and imaging with what the algorithm expects or imaging and radiotherapy provides fluent in, the research that underpins these anticipates being there? The issue of false profession-specific materials that can tools. A useful analogy is the Boeing 737 negative diagnoses, when AI reconstruction be used for values-based practice tragedies; a modification was made to the algorithms are used, is a growing concern (VBP) for AI. The education and training aircraft that allowed pilots to “feel” like they and one that requires further research of radiographers in VBP, facilitates were flying a 737, unaware that a sensor before these interventions are deployed engagement in shared evidence-based had been changed. What safeguards are in into routine clinical practice. decision-making with their patients, using place to ensure that scan planning, image dialogue about values (SCoR, 2018). This acquisition, or contouring in radiotherapy, Public and patient involvement (PPI) connection to our values, our patients, and is appropriate? These technologies are or patient, public and practitioner their needs, is key to how we continue to be used to enhance clinical decision-making. partnerships (PPPP) within medical the trusted professionals, and to be trusted imaging and radiotherapy (SCoR, 2018) other professionals and decision-makers, As radiographers we need to be improved should underpin all we do, including the by in the era of AI in healthcare. on improving the health of all, reducing development and implementation of AI not reinforcing healthcare inequalities. into our practices. PPI highlights the As diagnostic and radiotherapy Algorithms used to predict and anticipate need for true patient involvement at the radiographers, the gatekeeper between imaging demand are trained on healthcare discussion and planning stage of any referring clinician and patient radiation records and previous experience. For service improvement. This is something dose, will we continue to ensure that example, algorithms can predict non- that has been emphasised by the EFRS imaging and radiotherapy is optimised? attendance for screening mammography in recent years. Patient representation It is likely, for the foreseeable future, that appointments. So, how should this can often be tokenistic, even if well AI will provide “suggestions” for protocol information be used – to “double book” intentioned, so moving to an advocacy optimisation, but the radiographer will be clinics to ensure maximum use of limited
Figure 1: The importance of partnerships in all developments related to clinical service delivery (College of Radiographers, 2020).
ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 57 CONTRIBUTING STAKEHOLDERS VIEWS
diagnostic capacity, or to further engage 382(26): 2582. with potential non-attenders? If the sole Mehra MR, Desai SS, Ruschitzka F, measure of imaging is efficiency and Patel AN (2020). RETRACTED: the numbers seen, is this reinforcing Hydroxychloroquine or chloroquine with existing inequality? Patients in hard or without a macrolide for treatment to reach groups, lower socioeconomic of COVID-19: a multinational registry status, those with mental health issues, analysis. The Lancet. have historically not engaged with health Nagendran M, Chen Y, Lovejoy CA, Gordon prevention and promotion; by delegating AC, Komorowski M, Harvey H, et al. decisions on attendance to an algorithm do (2020). Artificial intelligence versus we, as radiographers, allow these biases clinicians: systematic review of design, and inequalities to be entrenched and reporting standards, and claims of reinforced? deep learning studies. BMJ, 368: m689. https://doi.org/10.1136/bmj.m689 It is also crucial that radiographers Society and College of Radiographers understand and can explain to patient (2018). Patient Public and Practitioner how their data could be used, benefits to Partnerships within Imaging and themselves and others, ensuring informed Radiotherapy: Guiding Principles. consent is obtained and compliance with www.sor.org/learning/document-library data protection laws ensured. Moving Society and College of Radiographers forward these concepts and experience (2018). Values-based Practice in will need to be embedded in radiographer Diagnostic & Therapeutic Radiography: education, from pre-registration training A Training Template. www.sor.org/ to postgraduate education and continued learning/document-library professional development. This is echoed by the EFRS and the ISRRT:
“The optimal integration of AI into medical safety, clinical imaging and radiation therapy can only be achieved through appropriate education of the current and future workforce and the active engagement of radiographers in AI advancements going forwards.” (ISRRT & EFRS, 2020)
References Harvey H (2017). Algorithms are the New Drugs. TowardsDataScience. com. Accessed: 24/07/2020. https:// towardsdatascience.com/algorithms- are-the-new-drugs-learning-lessons- from-big-pharma-997e1e7e297b International Society of Radiographers and Radiological Technologists (ISRRT), European Federation of Radiographer Societies (EFRS) (2020). Artificial Intelligence and the Radiographer/ Radiological Technologist Profession: A joint statement of the ISRRT and the EFRS. Radiography, 26(2): 93-5. www.radiographyonline.com/article/ S1078-8174(20)30037-7/fulltext Mehra MR, Desai SS, Kuy S, Henry TD, Patel AN (2020). Retraction: Cardiovascular Disease, Drug Therapy, and Mortality in Covid-19. N Engl J Med,
58 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 ISRRT SPECIAL EDITION ON WORLD RADIOGRAPHY DAY 2020 59