Guideline for the Use of Image Compression in Diagnostic Imaging Guideline for the Use of Image Compression in Diagnostic Imaging
Clinical Radiology
Guideline
Name of document and version: Guideline for the Use of Image Compression in Diagnostic Imaging, Version 2
Approved by: Faculty of Clinical Radiology Council
Date of approval: 13 August 2020
ABN 37 000 029 863 Copyright for this publication rests with The Royal Australian and New Zealand College of Radiologists ®
The Royal Australian and New Zealand College of Radiologists Level 9, 51 Druitt Street Sydney NSW 2000 Australia
New Zealand Office: Floor 6, 142 Lambton Quay, Wellington 6011, New Zealand
Email: [email protected] Website: www.ranzcr.com Telephone: +61 2 9268 9777
Disclaimer: The information provided in this document is of a general nature only and is not intended as a substitute for medical or legal advice. It is designed to support, not replace, the relationship that exists between a patient and his/her doctor. TABLE OF CONTENTS
1. Introduction 3
2. Data Compression in Medical Imaging 4 3. Appropriate Application of Data Compression 4 4. Recommendations 6 5. Related documents 7 August 2020 6. Acknowledgements 7 7. References 7 lege of Radiologists® | | © | The Australian Royal and New Zealand Col
Guideline for the Use of Image Compression in Diagnostic Imaging, Version 2
Page 3 of 8 Guideline for the Use of Image Compression in Diagnostic Imaging, Version 2 | © The Royal Australian and New Zealand College of Radiologists® | August 2020 compression algorithms to reduce Advances intechnolo procedure, and minimise way, and for such ape acquired, human resources, infrastructure and time inv imaging d that is increasin Medical 1. members. radiation oncology andmakes explicit the standards of ethical conduct the College expects of its The C of Ethics Code professional and organisational levels. Exemplified through strong leadership that is accountable to members; patient engagement at Accountability and collaborative attitude andpatient Exemplified through anethical approach: doingwhat isright, not wha Acting Integrity with high quality care; an attitude of compassion and empathy. Exemplified through anevidence Commitment toBest P Our Values services for optimum health outcomes by leading, training and sustaining our professionals. To drive the appropriate, proper and safe useof radiological and radiation oncological medical Mission Our b RANZCR the as peak group driving best practice in clinical radiology and radiation oncology for the Our Vision Australian Medical Council and the Medical Council of Zealand. New The work of the Colle radiology for the betterment of the ofpeople Australia and New Zealand. continuously improving the standards of training andpractice in diagnostic interventional and The Faculty of Clinical Radiolog radiation oncology in Australia and New Zealand. science and practice of the medical specialties of clinical radiology (diagnostic and interventional) and association of members who The RoyalAustralian and New Zealand College of Radiologists (RANZCR)is a not About College the enefit of our patients. ode defines the values and principles that underpin the best practice of clinical radiology and INTRODUCTION imaging isintegra ata isacquired at su
ch information should
gly relied upon inthe clinical manage
ge is scrutinised and externally accredited against industry standard by the gy ha
riod oftim
ractice t he riskof l totheprovision of mode ve created the opportunity some costtothecommunity deliver skills, knowledge, insight, time andcommit to promot
y, RANZCR, is the peak bi e, as - based culture, a focus onpatient outcomes and equity of access to not be the filesize unn - centric focus. t
o maximise theo maximise value gained fro ecessar discard olved inthe acquisition Page of eachimageinan y rep ed lightly
4 of8
rn medic eat tests. ment ofpatientsandtreat f or in termsof radiation exposure radi , andindeed should be pre
- national body for setting, promoting and al care ology systems to provides a ; itprovides wealth of effort of
t is expedient; a forward thinking m m theinitial
edical images.edical Once to partially use complex ment plannin - imaging for offset served insucha - , andthe profit ing the inf Medical g. Medical
ormation the increase in data volume created by new or more complex modalities. The purpose of this document is to provide guidance to radiologists about the appropriate levels of compression medical imaging practices may use.
2. DATA COMPRESSION IN MEDICAL IMAGING
Considerable effort has been expended in the wider community to produce destructive (irreversible or lossy) methods of image compression that still allow for the convenient transmission of large datasets with minimal appreciable loss of data fidelity. Examples include MP3 music files, JPEG photos and MPEG movies. August 2020
Broadly speaking, there are currently two main groups of image compression options available in medical imaging: lossless and lossy compression.
Lossless, or reversible compression is intended to reduce the size of the original image data set and so speed up image transmission and reduce required data storage space1,2. The image obtained after compression and then decompression is identical to the original image1,2. Typical compression ratios lege of Radiologists® | achieved range from 1.5:1 to 3.6:1.
Lossy, or irreversible compression techniques use algorithms which can compress images at much higher compression ratios than are achievable using lossless compression, resulting in faster image transmission speeds and smaller image storage space requirements1. With these techniques, the regenerated image is not guaranteed to be identical to the original image, as certain elements may have been removed when reducing the image size1,2; that is, some data are lost during the compression process, and some distortion may occur when the image is decompressed. Typical compression ratios achieved range from 5:1 to 50:1.
In the wider community, lossy data compression is often acceptable. Data from medical imaging examinations, however, may be put to many different uses, with potentially different requirements for fidelity: | © | The Australian Royal and New Zealand Col
• Extensive post-processing of large datasets (both from CT & MRI) is commonplace and will become more so with the wider deployment of ‘artificial intelligence’ techniques. • Semi-automated analysis of large datasets is commonplace and often necessary (e.g. breast MRI) • Automated follow-up for lung nodules and other pathology requires repeat access to the complete 3D dataset • Accurate, serial studies are often key to appropriate clinical management decisions
A loss of clinical data either compromises or has the potential to compromise the value of an imaging examination to a patient.
3. APPROPRIATE APPLICATION OF DATA COMPRESSION
3.1 Inconsistency in approaches internationally A number of studies overseas1,2,3,4 have examined the issue of lossy compression in medical imaging. There are numerous examples of studies testing the acceptable limits of image compression ratios, for many different modalities. While a general conclusion could be drawn that some levels of lossy compression are suitable for some purposes and some modalities (“Diagnostically Acceptable Image Compression”1), there remains considerable uncertainty as to exactly what level and type of compression is enough, or too much, for any particular examination or modality. For example, while compression of digital mammograms is not permitted in the USA by the Food and Drug Administration, the Royal College of Radiologists, Guideline for the Use of Image Compression in Diagnostic Imaging, Version 2 and the German Radiology Society in Europe, have published acceptable lossy compression ratios of 20:1 and 15:1, respectively.
Page 5 of 8 The European Society for Radiology collated guidelines for “Diagnostically Acceptable Image Compression” in its position paper of 20111. The RCR has subsequently withdrawn, and not replaced, its guidance on image compression; its recommendations were very similar to those of the German Roentgen Society4. The Canadian2 and German guidance remains current, as per an RCR guideline of 2008, (quoted in references (2) and (7), but subsequently withdrawn in May 2018); In general, the Canadian guidance permits higher levels of compression.
There remains considerable uncertainty as to the best metric for describing the quality of lossy compression. The compression ratio is widely used because it is readily available and can be August 2020 used to directly assess the effect of compression on storage requirements. However, it correlates poorly with measures of image quality. Other measures, such as ‘Peak Signal to Noise Ratio’ and ‘Structural Similarity’ have been tested, but there is no consensus on the most useful tool (see discussion in section 1).
A European review1 has noted that multiple cycles of lossy compression and compression have the potential to cause cumulative degradation of the image, especially if different algorithms are used, and therefore recommends against this practice.
3.2 Emerging technology
The majority of studies to date have concentrated on 2D image compression, since the commonly available (and commercially implemented) algorithms are thus focused. It has, however, been shown that new algorithms are required to best deal with the particular needs of 3D datasets5. It has also been shown that lossy compression produces visible difference in 3D images at relatively low compression ratios, though it is as yet unclear whether these have adverse effects on diagnostic performance1,6. These often large 3D datasets are no longer restricted to CT, but are commonly generated with MRI, and also, more recently, ultrasound. The effects of 2D image compression on 3D datasets, and how such effects may affect subsequent automated or semi-automated follow-up, have yet to be carefully examined in the literature. | © | The Australian Royal and New Zealand College of Radiologists® |
Compression of very thin section CT images has been shown to cause artefacts more frequently than that of images reconstructed at 5 mm or greater thickness; there is no data on the effects on Radiotherapy (RT) planning images. Given, however, that treatment planning scans for RT are usually thick-section reconstructions, and of limited number, some may argue that these are small enough datasets to be stored uncompressed.
3.3 Preserving clinical data for the future
It is well recognised that the availability of previous imaging studies often decisively influences the interpretation of a new study, by allowing the identification of changes in the findings, and an estimate of the rate of any such change.
Further, there is the potential in the future for new techniques to make use of data obtained and stored today in ways that are not currently possible. New techniques, new contexts and new insights will inevitably develop. Information could then be derived that was not immediately apparent at the time of the initial assessment. The methods of viewing and interpretation might f Image Compression in Diagnostic Imaging, Version 2 be very different to current practice. Stored images have the potential to be used in the future for more than simple visual comparisons.
In order to maximise the usefulness of lossy compression, national and international bodies have strongly recommended the use of lossy compression algorithms defined in the DICOM standard, both to maximise interoperability with other applications in the present, and to minimise the risk of the compressed images becoming orphan data in the future1,2,3. Guideline for the Use o Both the real value of the availability of historical images for comparison, and the hypothetical value of such potential future uses of losslessly compressed data, must be set against the real and immediate costs of storing and maintaining such data.
Page 6 of 8 3.4 Medico-legal issues
Legal reviews of lossy compression have been conducted by the CAR, and concluded that the use of such compression would not increase the liability of physicians, if compression were used and implemented appropriately2. However, this conclusion leaves much room for interpretation, and may lead to inconsistent application of compression. As demonstrated in section 3.1. above, what is considered by some as “appropriate” may not be seen as such by others. August 2020 3.5 Barriers to using lossless compression
It is acknowledged that there are constraints that may arise when using lossless compression techniques.
3.5.1 Bandwidth lege of Radiologists® | Bandwidth issues in parts of Australia and New Zealand may hinder transmission of an uncompressed or lossless compressed image. In these cases, the radiologist may make judicious use of appropriate levels of lossy image compression to speed the initial arrival of the images, with the patient’s interests foremost in their consideration.
If this approach is required, the practice should make the original images available for subsequent download as required. Poor bandwidth should not lead to the provision of a sub- optimal image data set to a patient or a referring practitioner, where the clinical context suggests that any data loss might put the patient at risk.
3.5.2 Storage costs
There is an argument that the costs of storing full data sets for every imaging examination
would be prohibitively expensive, and will become more so as imaging modalities become more © | The Australian Royal and New Zealand Col complex, powerful and productive. The CAR and RCR consider this ample justification for introducing lossy (irreversible) compression technique2,7. However, it is noted that, at least in the past, costs of digital data storage have fallen steadily.
3.5.3 Record retention
The RANZCR Standards of Practice under Sections 1 and 78 address the retention of a medical record. Public and private radiology clinics, practices or hospital departments across Australia and New Zealand should defer to this section of the RANZCR Standards of Practice in conjunction with any State or local legislation regarding disposal/retention/archiving.
4. RECOMMENDATIONS
Image compression is a useful tool in data transmission and storage, when used appropriately. In medical imaging, the level of compression should suit the needs of the examination, ensuring that the value of the examination is not adversely affected. It remains the responsibility of the Radiologist to ensure that archived and interpreted images are of diagnostic quality, and appropriate to the purpose of the original request. The RANZCR recognises that in the future, as technological advances are made, these recommendations may require modification.
• The RANZCR recommends the use of lossless compression where possible. However, where necessitated by infrastructure or cost barriers, conservative levels of lossy
compression may be used. If more aggressive lossy compression is used, compression levels Guideline for the Use of Image Compression in Diagnostic Imaging, Version 2 should generally be within the guidelines adopted by the CAR2. • The compression algorithm should be one defined in the DICOM standard. • Where higher compression ratios are used, these must be justifiable on the basis of prior published trial evidence that such ratios do not result in clinically significant data loss. Where 3-D very thin (< 1 mm) slice imaging is performed, data compression may adversely impact
Page 7 of 8 the quality of reconstructed images, and/or other image processing techniques, hence compression of such data (including its retention only in the form of thicker-section reconstructions) should be used with caution. • In future, if the cost of data storage continues to fall, and the role of advanced image processing techniques (including machine learning) expands, there may be stricter limitations on the use of lossy compression.
5. RELATED DOCUMENTS
August 2020 • Standards of Practice for Clinical Radiology
6. ACKNOWLEDGEMENTS
The Faculty of Clinical Radiology acknowledges the work by the eHealth reference group and the Safety, Quality and Standards Committee to update the guidelines
7. REFERENCES 1. European Society of Radiology. Usability of irreversible image compression in radiological imaging. A position paper by the European Society of Radiology. Ins Imag 2011; 2:103-115. Accessible at: https://insightsimaging.springeropen.com/track/pdf/10.1007/s13244-011- 0071-x 2. Canadian Association of Radiologists (CAR). CAR Standards for irreversible compression in digital diagnostic imaging within radiology. Ottawa. June 2011; https://car.ca/wp- content/uploads/Compression-in-Digital-Imaging-2011. 3. American College of Radiology. ACR-AAPM-SIIM Technical Standard for Electronic Practice of Medical Imaging. ACR Practice Guideline. Reston VA. 2017 https://www.acr.org/- /media/ACR/Files/Practice-Parameters/Elec-Practice-MedImag.pdf 4. Loose, R (R); Braunschweig, R (R); Kotter, E (E); Mildenberger, P (P); Simmler, R (R); | © | The Australian Royal and New Zealand College of Radiologists® |
Wucherer, M (M) Compression of digital images in radiology - results of a consensus conference, RöFo: Fortschritte auf dem Gebiete der Röntgenstrahlen und der Nuklearmedizin (Rofo), published in Germany, 2009-Jan; vol 181 (issue 1) : pp 32-7 5. Kim, B et al. JPEG2000 3D compression vs. 2D compression: an assessment of artifact amount and computing time in compressing thin-section abdomen CT images. Medical Physics 2009; Mar; 36 (3): pp835-844 6. Erickson BJ, Krupinski E, and Andriole KP. A multi-center observer performance study of 3D JPEG compression of thin-slice CT. J Digit Imaging 2010(Oct) 23(5):639-43. 7. Royal College of Radiologists. Guidelines and standards for the implementation of new PACS/RIS solutions in the UK. London, Royal College of Radiologists, 2011. BFCR(11)4 Royal College of Radiologists, June 2011. https://www.rcr.ac.uk/sites/default/files/docs/radiology/pdf/BFCR%2811%294_PACS.pdf 8. RANZCR. RANZCR Standards of Practice for Diagnostic and Interventional Radiology, v10.2, 2017; www.ranzcr.edu.au f Image Compression in Diagnostic Imaging, Version 2 Guideline for the Use o
Page 8 of 8