A Geriatric Suite of Medical Applications for Android Powered Devices Frank A

Florida State University Libraries

Electronic Theses, Treatises and Dissertations The Graduate School

2012 A Geriatric Suite of Medical Applications for Android Powered Devices Frank A. Sposaro

Follow this and additional works at the FSU Digital Library. For more information, please contact [email protected] THE FLORIDA STATE UNIVERSITY

COLLEGE OF ARTS AND SCIENCES

A GERIATRIC SUITE OF MEDICAL APPLICATIONS FOR ANDROID POWERED DEVICES

FRANK A. SPOSARO

A Thesis submitted to the Department of Computer Science in partial fulﬁllment of the requirements for the degree of Master of Science

Degree Awarded: Spring Semester, 2012 Frank A. Sposaro defended this thesis on March 26, 2012. The members of the supervisory committee were:

Gary Tyson Professor Directing Thesis

Zhenghao Zhang Committee Member

Zhenhai Duan Committee Member

The Graduate School has veriﬁed and approved the above-named committee members, and certiﬁes that the thesis has been approved in accordance with the university requirements.

ii To my family, one of us had to ﬁnish college. To my uncle Vince, I miss you deeply. To the many that overlooked, used, and abused me, you see me? Hi hater. Thanks for the motivation.

iii ACKNOWLEDGMENTS

There is a time in everyones life when they begin their journey with absolutely no knowledge of what is in store. This is the point where the journey can end before it even begins. I would like to thank the man who introduced me to programming, the journey which has become my entire life. Mr. Joseph A. Tosh of Atlantic City High School. I would also like to thank my adviser, mentor, and friend Dr. Gary Tyson. He has let me forge my path and both create and conquer my own challenges. The experience I have gained starting The Mobile Lab @ FSU will follow me for the rest of my career. My time at FSU changed me in ways I would have never imagined. Lets go Noles.

iv TABLE OF CONTENTS

ListofTables...... viii ListofFigures ...... ix Abstract...... x

1 Introduction 1 1.1 Overview ...... 1 1.2 Motivation ...... 1 1.2.1 Aging of the Baby Boom Generation ...... 1 1.2.2 Increased Life Expectancy ...... 1 1.2.3 Proliferating Chronic Conditions ...... 3 1.2.4 Incident and Cost of Chronic Disease in the US ...... 3 1.2.5 Sophisticated Technology Targeted at Older Adults ...... 4 1.3 TargetUsers ...... 4 1.3.1 OlderAdults ...... 4 1.3.2 ConcernedParties ...... 4 1.3.3 Inﬂuencers ...... 4 1.4 CurrentSolutions...... 5 1.4.1 JitterBug ...... 5 1.4.2 LifeAlert ...... 5 1.4.3 LifeLink ...... 6 1.4.4 AreaSummary...... 7

2 iFall: A Fall Monitoring Application 8 2.1 Overview ...... 8 2.2 Introduction...... 8 2.3 MaterialsandMethods...... 9 2.3.1 Hardware ...... 9 2.3.2 Software...... 9 2.3.3 Fall Detection ...... 10 2.3.4 Application Features ...... 11 2.4 Challenges...... 11

3 iWander: An Android Application for Dementia Patients 15 3.1 Overview ...... 15 3.2 Introduction...... 15 3.3 MethodsandDesign ...... 16 3.3.1 Software ...... 16 3.3.2 Wander Detection ...... 16 3.3.3 Alert Actions ...... 17 3.3.4 Automotive Travel ...... 19 3.4 Challenges...... 19

v 3.4.1 Hardware ...... 19 3.4.2 Patient Interaction ...... 20 3.4.3 Implementation ...... 20

4 BEAT: Bio-Enviormental Android Tracking 21 4.1 Overview ...... 21 4.2 Motivation ...... 21 4.2.1 Heart Disease ...... 21 4.3 CurrentSystems ...... 22 4.3.1 Wireless Body Area Networks ...... 22 4.3.2 m-Health ...... 22 4.3.3 Fitness...... 22 4.4 BEATDesign...... 22 4.4.1 The choice of mobile operating system ...... 22 4.4.2 Real-time monitoring ...... 23 4.4.3 Emergency response ...... 23 4.4.4 Contextual data analyses and user interactions ...... 23 4.4.5 Long-term analysis ...... 23 4.4.6 Power management ...... 23 4.5 Implementation...... 24 4.5.1 Hardware ...... 24 4.5.2 Software ...... 24 4.6 Evaluation...... 25 4.6.1 Storage ...... 25 4.6.2 PowerOverhead ...... 25

5 Conclusion 27 5.1 Applications...... 27 5.1.1 iFall...... 27 5.1.2 iWander...... 27 5.1.3 BEAT...... 27 5.1.4 Alert...... 27 5.2 CurrentState...... 28 5.2.1 Human Subject Testing ...... 28 5.2.2 Real World Evaluation ...... 28 5.2.3 APIs, Extendability, and Licensing ...... 28 5.3 FutureWorks...... 29 5.3.1 Additional Accelerometer Sensors ...... 29 5.3.2 Activity Classiﬁcation ...... 29 5.3.3 Daily Medical Monitoring ...... 29 5.3.4 Depression ...... 29 5.3.5 Server Data Analysis ...... 29 5.3.6 Vision Aid ...... 30

A Documents 31 A.1 Human Subjects Testing Approval ...... 32 A.1.1 IRBContinued ...... 33 A.2 HumanSubjectsTestingRenewal...... 34 A.3 SupportforTesting...... 35 A.4 TestingConsentForm ...... 36 A.4.1 Consent Continued ...... 37 Bibliography ...... 38

vi BiographicalSketch ...... 43

vii LIST OF TABLES

1.1 Incident and Cost of Chronic Diseases in the US ...... 3

4.1 Projected number of data points for each sensor ...... 25

4.2 Size estimation of data from various sensors (KB) ...... 26

viii LIST OF FIGURES

1.1 Percentage of population in United States that are over-under 65 years of age. . . . . 2

2.1 Total gravity readings of a typical fall ...... 10

2.2 Total gravity readings while running ...... 13

2.3 Total gravity readings while sitting then standing ...... 13

2.4 Total gravity readings when answering the phone ...... 14

3.1 Bayesian Network Variable Relations Aﬀecting Wandering Probability ...... 18

4.1 BEATComponentOverview ...... 26

ix ABSTRACT

The rising number of senior citizens is approaching an all time high. With this comes a rising number of common chronic conditions where treatment is costly and time consuming. Frequently these conditions make it unsafe for older adults to live independently. This creates a burden on loved ones because situations arise which require assistance. Slip & falls, wandering, and daily health monitoring are the top burdens loved ones face with an aging adult. Today, however, these burdens can be minimized by using smartphone technology. Modern devices are capable of automatically recognizing, reporting, and remembering these situations. The proposed system is a collection of several applications that enable this functionality for Android powered devices. These applications focus on monitoring falls, wandering, and storing health information in the users daily life. The software achieves these tasks by gathering and analyzing data from various sensors both on and off the device. Several algorithms are applied to monitor and report dangerous events. The algorithms range from learning networks to timing based thresholds. If a dangerous event is detected it can be easily canceled by the user in order to reduce false positives. After confirmation, or lack of, the user’s loved ones are promptly notified to further assess the situation. By using Social Monitoring, false positive to costly emergency medical professionals are kept to the bare minimum. Thus eliminating the need for paid monitoring services. The system also provides an API to allow for other developers to integrate with the event analysis. Once in place, a wealth of information can be recorded and used to help identify further problems and solutions.

x CHAPTER 1

INTRODUCTION

1.1 Overview

The geriatric suite is a collection of several medical related applications geared for older adults, typically 65+ years of age. Speciﬁcally targeting those who are at risk due to the chronic conditions associated with growing older. They enable users to live a safer, independent life while providing peace of mind to loved ones. All the concerned parties can rest assured knowing that they will automatically be alerted in a real emergency. The various applications are designed to minimize diﬀerent risks an older adult faces, such as falling or wandering. If a problem is detected, the application will invoke a special alert sequence to notify concerned parties while reducing the rate of false positives. The suite was created to run on the Android powered devices and provide services to third party applications.

1.2 Motivation

The need for a suite like this is driven by several demographic trends including: the aging of the baby boom generation, increased life expectancy, proliferating chronic conditions, and sophisticated technology targeting older adults. As time continues it is becoming more feasible to use smart phone technology in order to solve these types of problems. Several vendors oﬀer services as a solution, however they tend to be limited and expensive. The proposed system is comparable to current systems while both extending functionality and reducing cost.

1.2.1 Aging of the Baby Boom Generation In January of 2007, the oldest baby boomers turned 62 and became eligible for social security. For the next sixteen years, a member of the baby boom generation, the 78 million Americans born between 1946 and 1964, will reach that age every 8 seconds [1]. And, as Figure 1.1 indicates, the percentage of the US population aged 65 and older is increasing signiﬁcantly from 5% (6.2 million people) in 1930 to a projected 20% (70.2 million) in 2030.

1.2.2 Increased Life Expectancy For the last ﬁfty years, life expectancy in the US has also been increasing. The U.S. Centers for Disease Control and Prevention recently reported that the life expectancy in the US reached almost 78 years in 2007 - a record high. According to the report, life expectancy has increased 1.4 years since 1997. The average lifespan in the U.S. increased by more than 30 years during the 20th century. [2]

1 Figure 1.1: Percentage of population in United States that are over-under 65 years of age.

2 Table 1.1: Incident and Cost of Chronic Diseases in the US Disease Incidence Future Incidence Cost Osteoporosis 10 Million 14 Million $14 Billion 2002 2009 2009 Low Bone Mass 32 Million 47 Million 2002 2020 Fractures 2 Million 2005 Alzheimer’s 5.1 Million (80% of all Dementia Cases) 450K new cases/yr 615K new cases/yr $148 Billion 2010 2029 Diabetes 12.2 Million 24.5 Million $174 Billion (Ages 60+) (Ages 60+) (All Ages) 2050 Parkinson’s 1 Million $25 Billion

1.2.3 Proliferating Chronic Conditions Although life expectancy has increased, the quality of life may have decreased due to the chronic medical conditions older adults face. Dr. David L. Katz, director of the Prevention Research Center at Yale University School of Medicine stated:

Modern medicine may help humans live relatively long lives ... There is much more to living well than not dying. One factor that contributes to the quality of life for older adults is the ability to live independently.

According to The Silver Book: Osteoporosis, ninety percent of the 65 and older population have at least one chronic condition. Many have multiple chronic conditions. These chronic conditions put older adults at risk and reduce their ability to live independently. In addition, they put an enormous burden on family members, other care givers and the economy as a whole. These chronic conditions include:

• Osteoporosis, which aﬀects people later in life with about 55% of the 50 and older population aﬀected by osteoporosis or low bone mass

• Dementia, which the largest percentage of which is Alzheimer’s Disease

• Diabetes, which any age group with diabetes could beneﬁt from medical applications integrated into Android, especially those who are insulin dependent

• Parkinson’s Disease, which generally aﬀects people over 60 and older

People with dementia, diabetes and Parkinsons disease are particularly at risk for falling. Falling is the 6th leading cause of death in ages 65+. It is also responsible for 40% of nursing home admissions. People with osteoporosis have weak bones that break easily as the condition progresses and falling is a major concern for them. [3, 4]

1.2.4 Incident and Cost of Chronic Disease in the US Table 1 displays a number of chronic diseases along with the number aﬀect and the amount spent for care. As the population ages, these number increase because the number of people in the 65+ age also increase.

3 1.2.5 Sophisticated Technology Targeted at Older Adults Gerontechnology is a new field of research that studies older adults and technology, particularly the technology gap. Older adults due to their attitudes and abilities are slower to adopt new forms of technology. They are afflicted with declining cognitive processes, decreased memory and difficulty maintaining attention. These afflictions make it more difficult for older adults to learn new skills. To address these issues and to reduce the technological gap, a number of companies have developed technologies that address the needs of older adults. These range from cell phones to computers, many of which provide background services to address the challenges older adults face in adopting new technologies. With the advent of the smart phone, the technological gap can be further reduced. Smart phones have significant processing power, large memory, graphics /video/touch screens, 3rd party applications, wireless 3g Internet and GPS, motion sensing, and compasses. Using Google’s Android operating system, developers can create their own interfaces and applications to address the needs of a specific user. The proposed geriatric suite utilizes background running processes, simple user interfaces, and automation in order to limit the amount of interaction necessary between the user and the application.

1.3 Target Users 1.3.1 Older Adults The end users are older adults whose risk of falling, wandering, and forgetfulness is increased due to common chronic conditions. Since baby boomers tend to be healthier than their parents, most will not need this system until they reach 65 or older (65+). Although technology poses a challenge for many who are 65+, the Pew Research Center found that the highest growth rates in broadband use are happening with the 65+ segment. They found that the Internet/broadband use for the 65+ increased from 19% in May 2008 to 30% in April 2009 . Use of cell phone by the 65+ segment is even higher with 46% using a cell phone [5]. The applications encourage this adoption by designing around the capabilities and challenges of the 65+ segment.

1.3.2 Concerned Parties Concerned parties include spouses, children, friends, caregivers and others who are worried about the well being of their elderly, loved one. Their goal is to attain peace of mind by ensuring that they will be automatically notified if their loved one falls, wanders off, forgets their pills, gets confused or needs general assistance. In many cases, these individuals will obtain and configure the applications for the end user. Thus, they are an important and significant in the design process and flow of information

1.3.3 Influencers Influencers are groups of individuals or organizations that will recommend the system to older adults if they perceive that the it delivers value by reducing risk and increasing quality of life. In addition, influencers may have use for wealth of bio-environmental data mined for the system. This data can be researched to learn trends and develop other treatments. Influencers include: physi- cians, family practitioners, gerontologists, neurologists, orthopedists and osteoporosis specialists Gerontology specialists. There are over 101 professions in gerontology including adult day program coordinators, nurses, social workers, art therapists, physical therapists, aging in place specialists, etc. Also organizations that serve older adults such as senior centers, hospices, AARP, day programs, state health organizations, etc. The influencers are an important distribution channel and will help

4 the adoption of the system. In most cases, older adults trust their doctors and other professionals and organizations that are improving their quality of life.

1.4 Current Solutions 1.4.1 JitterBug Jitterbug Wireless, is an American mobile virtual network operator founded by Arlene Harris and Martin Cooper in 2006 and based in Del Mar, California. The brand name Jitterbug refers to both the handsets and the wireless service the company provides. Jitterbugs trademark is a simplified mobile phone that is targeted at baby boomers and older Americans. [?] Jitterbug offers airtime on a month-to-month or prepaid basis. They currently offer three mobile phone models, all of which are clam shells. The Jitterbug and The Jitterbug in Graphite have large back-lit buttons to make and receive calls. The Jitterbug One-Touch model is a simpler phone with three buttons programmed to contact an operator, a contact designated by the user, and 9-1-1 emergency services. All phones have speakers with increased volume levels, hearing aid compatibility and brighter screens with larger text than on most cell phones. Phones emit a dial tone when opened to simulate a land line phones behavior. The menu system uses a question- based interface with keys labeled YES and NO. Jitterbug subscribers can reach a live operator 24 hours a day. Operators greet customers by name when they call and are trained to connect calls, provide directory assistance and make changes and updates to customer phone number contact lists. Calls to an operator deduct minutes from the customers monthly allowance. In May 2009, Samsung recalled some phones manufactured for Jitterbug that could not connect to 9-1-1 when in a no service area. Additional Services. Voicemail, voice-activated dialing, and call history can be turned on or off. Customers can pay only for what they need from a set of options. The Jitterbug LiveNurse service gives Jitterbug owners 24 hour access to live nurses for tailored health and wellness advice. For a small monthly fee of $4 a month a subscribers can call directly from Jitterbug phone and the only additional cost is the airtime they use. In addition, subscribers also get free access to an audio health library for prerecorded health information. The Jitterbug LiveNurse service is provided for Jitterbug by FoneMed, a company based in Colorado Springs, CO. Their nurses are located in the US and Canada. Many of them are in New Newfoundland, Canada and are specially trained to understand the customers. Criticisms. Jitterbug boosts several strengths geared towards target users. Simple design with no confusing icons make it easy to use. Navigation is and functionality are basic yes or no buttons. Soft ear cushion, powerful speaker, big, back-lit buttons and big, bright text are also a drawing point. Jitterbug is one of the most expensive prepaid plans on the market, both in terms of per-minute airtime cost, and minimum monthly cost (up to 13 times more expensive per month than T-Mobile prepaid, and 4 times as expensive as PagePlus, a Verizon mobile virtual network operator). Also, minutes expire after 60 days, rather than 365 or 120 for T-Mobile and PagePlus, respectively. The phones themselves are also quite expensive, generally around $147, despite being ultra low-end models. There is also a $35 activation fee, which is unusual for a prepaid service. The Jitterbug also lacks the advanced hardware and software functionality. Even a simple functionality of Bluetooth is not available with these phones

1.4.2 Life Alert Life Alert Emergency Response, Inc. is a nationwide company, with headquarters in Encino, California, which provides services that help the elderly contact emergency services. The company’s

5 system is based around a main unit and a small wireless help button that is worn on the user at all times. Life Alert was founded by Isaac Shepher in 1987. Former Surgeon General Dr. C. Everett Koop has appeared in commercials for Life Alert since 1992, stating that he wears one. Life Alert service consists of a pendant-shaped device, worn on a necklace or wristband, an automated dialer connected to a telephone line, and an emergency dispatcher at the other end of the line. When an elderly or handicapped person falls down and cannot get up, or has an emergency, such as a break-in, a telephone may be out of reach. To get quick assistance, the customer simply presses a button on the pendant. This activates the automated dialer and calls a company call center. The dialer then works as a speaker phone. At the other end is an emergency services dispatcher to alert the authorities to the customer’s predicament. The service is marketed as a way for seniors to continue living at home rather than move to a retirement home and is available nationwide.

Services.

• Medical Protection Life Alert members are protected against dire situations. Life Alert provides vital protection 24/7 in the event of any medical emergency, including heart attacks, stroke and falls.

• Monitored Fire Protection Life Alert offers its members an optional added benefit that enhances our home monitoring service: an advanced smoke detection and response system, monitored 24/7 by live dispatchers. Whether you are home or not, Life Alert can send help any time of day or night if fire or smoke is detected.

• Third Layer Protection Protection away from home : Life Alert can provide its customers with a special 911 cell phone as a part of their protection package. All the customers have to do is to push one large, easily-accessible button to reach a local 911 operator. This has got Nationwide coverage. No extra contract or service fee or even roaming fees is charged.

• CO Gas Poisoning Protection Another optional beneﬁt oﬀered by Life Alert to its members is its Carbon Monoxide (CO) gas detection and response system, which further enhances our home monitoring service. This CO gas system, like our smoke alarm system, is monitored 24/7 by live dispatchers. Whether you are home or not, Life Alert can send help any time of day or night if CO gas is detected.

Criticisms. Very poor customer service. There are a number of negative reviews on the customer service of the life alert customers. The customer service oﬃcer is neither approachable nor helpful, if at all connected. They even charge for the customer service. [6] On an average life alert charges about $80 per month per customer. This is the basic package. If the other features like CO detection, Fire alarms and 911 service are included, then the price may go up to $120 per month. This makes this company very pricey to go with.

1.4.3 Life Link LifeLink is based of the PERS technology. The following is an except from their brochure.

LifeLink works anywhere there is a phone line. In the event of an emergency, the system calls up to four phone numbers that are programmed during the setup process; local numbers, long distance numbers, also 911. It runs on regular house power and uses a 9 volt battery as a backup in case of power failure. When there is an emergency, the wireless pendant signals the Console to begin the emergency calling process.[7]

6 Criticisms. This system is much like Life Alert’s. However, LifeLink does not oﬀer a monitoring service. It uses the concept of social monitoring, discussed in the technology section. Due to this, LifeLink customers do not have to sign a contract or pay for monthly service. The one time pay is a drawing point for many customers. This system suﬀers from the same downfalls as all PER systems. The user must be in physical range of the receiver. The user must also initiated a call for help. In addition, with out a monitoring service the user may not be attended to if all four social contacts are not available.

1.4.4 Area Summary Jitterbug. The Jitterbug is aimed to serve the elderly with large buttons and Yes/No navigation. It uses mobile virtual network operator (MVNO) technology and with several calling plans, including roll over minutes. MVNO technology enables Jitterbug to use existing infrastructure and resell minutes bought in bulk. It has become popular due to its big buttons and bright colors provided for users with impaired vision. The initial price is about $150 with monthly costs of $15-$80. This averages roughly a minute, which is significantly higher than the average. Jitterbug provides additional monitoring services such as Live Nurse. This service allows users to speak with a registered nurse with a push of a button. The major strengths of Jitterbug is the LiveNurse service, yes/no navigation, and high contrast, bold interface. A major drawback is the phone’s lack of versatility. It is not easily programmable and is not greatly personalized for the users needs. Drawbacks also include high price and lack new technology. Life Alert. Life Alert promises peace of mind to its customers using PERS Technology. It has water proof push buttons offered as pendants or watches. When the button is pressed, a call is connected to a Life Alert operator through a receiver. Additional services such as CO monitoring, fire alarm, and intruder detection are offered. A major weakness is price. Life Alert requires a $200 start up fee along with $60 a month. Also Life Alert customers often complain about poor customer service. The receiver range is poor too. Oftentimes the receiver can not detect a pushed button when the person is in another room. However, the biggest weakness is manual activation of calls. The user must push the button to call for help. It cannot detect a fall automatically, which becomes useless if the user losses consciousness after a fall. LifeLink. The LifeLink system uses a PERS similar to Life Alert. However, a 24 hour monitoring service is not attached. Activating the product will contact a predetermined list of five phone numbers. These numbers are programmed at time of purchase and can be anything ranging from a social contact, local police, or doctor. The strength of this product is the low cost. It consists of a one time purchase of $200. Like Life Alert, manual activation and versatility are the major weaknesses. All of the evaluated technologies offer valuable services, but have clear weaknesses. The proposed system attempts to take the highlights of each technology and unify them under one platform. It’s specifically designed to be useful and appealing to the targeted users. The following chapters build the case for each single application that makes up the entire geriatric suite.

7 CHAPTER 2

IFALL: A FALL MONITORING APPLICATION

2.1 Overview

Injuries due to falls are among the leading causes of hospitalization in elderly persons, often resulting in a rapid decline in functionality and death. Rapid response can improve the patients outcome, but this is often lacking when the injured person lives alone and the nature of the injury complicates calling for help. This paper presents an alert system for fall detection using common commercially available electronic devices to both detect the fall and alert authorities. We use a common Android-based smart phone with an integrated triaxial accelerometer. Data from the accelerometer is evaluated with several threshold based algorithms and position data to determine a fall. The threshold is adaptive based on user provided parameters such as: height, weight, and level of activity. These variables also adapt to the unique movements that a cellphone experiences as opposed to similar system which require users to mount accelerometers to their chest or trunk. If a fall is suspected a notification is raised requiring the user’s response. If the user does not respond, the system alerts prespecified, social contacts with an informational message via SMS. When a contact responds with an incoming call the system commits an audible notification, automatically answers the call, and enables speakerphone. If a social contact confirms a fall, an appropriate emergency service is alerted. Our system provides a realizable, cost effective solution to fall detection using a simple graphical interface while not overwhelming the user with uncomfortable sensors.

2.2 Introduction

As age related changes in reaction time and balance reduce the capabilities of people, the likely hood of a fall leading to signiﬁcant injury increases. Not only are fall related injuries the number one reason for emergency room visits, it is also the leading cause of injury-related deaths among adults 65 years old and older [8]. Every year, more than 11 million people fall [9]. In 2005, unintentional falls accounted for an estimated 56,423 hospitalizations and 7,946 related deaths in the United States [10]. Many of these deaths are a result of a “long-lie,” an extended period of time where the victim remains immobile on the ground [11]. Just the simple fear of a long-lie or falling can lead to one’s lower mental health, isolation, and general degradation of quality of living [12, 13]. Current systems are available that attempt to reduce the long-lie period by alerting emergency services when a fall has been detected. These systems commonly use one of three methods for classifying a fall:

1. Acoustic/vibration recognition: This is achieved by having a device, usually implanted in the ﬂoor, monitor sound and other vibrations. It listens for the vibratory signature of a human fall, which is vastly diﬀerent from the signatures of walking, small objects falling, and other common activities [14, 15].

8 2. Image recognition: By using overhead cameras in a ﬁxed location, one can track objects and learn movement patterns. The system adapts to the locations in which a single human enters/exits the room and remains inactive (lying/sitting on bed/chair). Common paths from entry points to inactive areas are then traced and remembered. It suspects a fall if a person becomes inactive in middle of a common path [16, 17, 18, 19].

3. Worn Devices: These systems require the user to wear external sensors. The devices track the vector forces exerted on the user. Usually these devices are a tri-axial accelerometer or gyroscope. If a speciﬁc pattern or threshold is broken, the device alerts a wireless receiver, which would then alerts emergency contacts [13, 20, 21].

The majority of fall detection systems require some application specific hardware and software design. This increases cost and limits the commercial viability to the wealthiest, or most impaired, users. Many also have significant installation and/or training times, also limiting greater adoption. Despite implementation differences, all designs have the same requirements: reliability, ease of installation/use, and restriction of false positives [13]. Falls are often sparse with months between occurrences, thus the system must always be ready and accurate. If installation costs or training time is high, users will reject the system. However, the major reason for failure is rejection by monitoring services due to a high number of false alarms [22, 23]. We propose a low priced system that is well suited to all the requirements by using existing mainstream technologies that are reliable and ubiquitous. Our approach is to use a worn device that billions of people already possess, a programmable cellular phone [24]. Using existing cell phone technology not only reduces the cost to the patient, it also exploits a greater range of communication capabilities and integrated hardware and software features. Touch screen response and voice recognition, common to smart phones, provide a reliable interface with the user. By using similar interfaces to applications the user frequently uses, the rare interaction with the fall detection software should be familiar. Cell phones are also more discrete than a dedicated monitor device, this hopes to reduce rejection due to the device’s poor aesthetic value and intrusiveness [13]. To limit false positives we implement several fall detection algorithms and two stages of communication. When a fall is detected, we first communicate with the user. If the user does not respond, we then attempt to contact members in his or her social network. If both fail or the social contact confirms a fall, the system alerts an emergency service.

2.3 Materials and Methods 2.3.1 Hardware The prototyped application is designed for the HTC G1. The G1 has a Qualcomm R MSM7201A running at 528 MHz. Its dimensions are 117.7 mm × 55.7 mm × 17.1 mm and weighs 158 grams. The touch screen has a 320 x 480 resolution. It has 192 MB of RAM and a 3.2 mega-pixel camera. It supports a 3G, Wideband Code Division Multiple network running at 2100 MHz, however our prototype is using the T-Mobile 2G network. The phone also supports sending and receiving SMS and MMS messages. A GPS receiver is also embedded and it is both 802.11g and Bluetooth R 2.0 capable. [25]

2.3.2 Software We chose to use the Android software stack produced by Google. Android is an open source framework designed for mobile devices. It packages an operating system, middle ware, and key applications [26]. The Android SDK provides libraries needed to interface with the hardware and make/deploy an Android application [27]. Applications are written in Java and run on the Dalvik virtual machine. Android uses a SQLite database to store persistent data.

9 Unlike dedicated systems, our software is intended to integrate with the phone’s existing applications. Our application, iFall, must share resources with the other apps. To make for a pleasant integration, iFall runs as inconspicuously as possible while using limited resources. We launch a background service that constantly listens to the accelerometer. Only when the algorithm described in the following section suspects a fall will the service wake up and interrupt the user. If the user responds, the previous activity is restored and iFall will sleep again. By only waking up the activity when a fall is suspected or requested by the user, we allow applications to run on top of iFall while we minimize our memory consumption.

2.3.3 Fall Detection Activities of Daily Living (ADL) are normal activities such as walking and standing. The forces exerted during ADL are usually different than the forces during a fall. By taking the root-sum- of-squares of the accelerometer’s three axles, we are able to determine the acceleration [11]. A fall must start with a short free fall period. This causes the acceleration’s amplitude to drop significantly below the 1G threshold [11]. This represents the period of time when the actual fall is taking place. The fall must stop and it causes a spike in the graph. The amplitude then crossing an upper threshold suggests a fall. Typically the minimum value for the upper threshold is around 3G [28]. If a person is seriously injured in a fall they usually remain on the ground for a period of time. This is characterized by the 1G flat line at the end of the graph. All of these events occur within a short duration. Figure 2.1 is a graph of a typical fall. If the amplitude crosses the lower and upper thresholds in the set duration period a fall is suspected. However, relying strictly on this method would produce an intolerable number of false positives since certain ADL and the upper threshold can overlap [29]. We refine the algorithm by taking position into consideration. The assumption is a fall can only start from an upright position and end in a horizontal position [30]. Thus the difference in position before and after the fall is close to 90 ◦ [20]. A fall is only suspected if both thresholds are crossed within a duration and the position is changed. Dropping the phone is a frequent motion that resembles a suspected fall. Also a fall may occur but, be minor leaving the user unharmed. To prevent these false alarms we add one more stage to the process, recovery. If a fall is suspected, we start a short timer. This timer allows a fallen user to regain an upright position or a dropped phone to be picked up. If the original position is resorted within the time limit the algorithm is reset. If the timer expires and position is not restored, we assume the phone/user is lying on the ground [31]. It then emits a prompt that requires the user to respond within a short time window. A fall is confirmed if the user does not respond. This allows users to reduce the number of false positives. An alert only sends when a fall is confirmed.

2.3.4 Application Features The iFall application is designed to be simple to use. To achieve this, we severely limit the number of buttons and options available to the user. The main screen consists of one button and a label. The button starts and stops the fall monitor while the label displays the state. The fall monitor is implemented as a low-powered, Android service. A service allows the fall monitor to constantly run the background. When the monitor suspects a fall, an intent is sent to iFall. This wakes up the application and attempts to get the user’s attention by repeatedly vibrating, ﬂashing LEDs, and playing an audio message. The app prompts the user with a simple pop-up window telling them to press an on-screen button if they are okay. Pressing the button cancels the alert, and the interrupted activity is restored. This gives users the opportunity to eliminate false positives [23, 21]. The iFall application has additional methods to reduce the number of false positives. We allow the amplitude’s upper threshold described in the ’Fall Detection’ section to be variable. The application displays a small list of conﬁguration options when the phone’s menu key is pressed. One option is

10 Figure 2.1: Total gravity readings of a typical fall

11 to adjust the sensitivity, the capacity to detect a fall [22]. So the less sensitive, the higher the upper threshold is. Given information such as age, weight, height, and level of activity are also factored into the equation [17], [20]. The other option under the application’s menu is Add a contact. This allows the user to add social contacts to their iFall, emergency contact list. Using social contacts to confirm a fall before alerting an emergency service is another method for filtering false positives. When a fall is confirmed, every contact in the iFall emergency list is sent a SMS message [32]. This message states that a fall was detected at the given time and includes the GPS coordinates of the fall. It also asks the contact to call the fallee. When called, a message is played on the fallee’s phone and the call is automatically answered and placed on speaker. Enabling bidirectional voice communication between the fallee and social contact reduces the number of false positives [23]. The dedicated emergency services are only notified when a social contact also confirms the fall, or in the case if no social contacts call the fallee.

2.4 Challenges

Using smart phone technology for fall detection has numerous advantages in cost and capability of the system. However, leveraging an existing system does pose challenges that single use detectors can avoid. One advantage of using a smart phone, is that the user is more likely to carry the phone throughout the day since it seen as indispensable in daily living, whereas users may forget to wear special micro sensors [33]. Unfortunately, it may be difficult to convince users to mount the phone to various body parts in order to improve fall detection rate [34]. Instead, the software must dynamically adjust to different methods of carrying the phone (e.g., in the purse, pants or shirt pocket, or on a belt or neck clip). This requires the software to classify acceleration parameters of general use to identify the correct parameters for the fall detection logic. To adapt for different carrying methods, we dynamically adjust the upper threshold and staring position. If the phone is carried on more accelerated body parts, such as the arms, the level of activity is automatically be risen. This causes the upper threshold to be greater [34]. Likewise, more stationary spots like the trunk will lower the threshold [11]. To account for the different orientations that phone may be held, like vertically or horizontally, we dynamically adjust the starting position. If the phone is resting for an extended period of time with 1G acceleration, we designate that to be the starting position. This allows the position to be dynamically set as the user interacts with the phone throughout the day. Figure 2.2 graphs the walking/running activity. Running’s amplitude can break the lower and upper threshold. If the user suddenly stops, it can cause an extended period of 1G acceleration. These events together suggest a fall. However, a prompt will not be given because the phone’s starting and ending position are the same. Figure 2.3 graphs the sitting and standing activity. This activity changes the phone’s position. However, sitting and standing’s acceleration will not usually break the upper threshold. Both experiments were performed while the phone was in the user’s front pants pocket. Some interactions with the phone, such as answering then ending a call, can break the thresholds and change position (see figure 2.4). Additional refinements to our algorithm must be made to prevent this. We do not allow the starting position to be dynamically changed if a call is in session. This will filter out the false positives in actions such as raising the phone to the user’s ear to start a call and lowering the phone from the user’s ear to end a call.

12 Figure 2.2: Total gravity readings while running

Figure 2.3: Total gravity readings while sitting then standing 13 Figure 2.4: Total gravity readings when answering the phone

14 CHAPTER 3

IWANDER: AN ANDROID APPLICATION FOR DEMENTIA PATIENTS

3.1 Overview

Non-pharmacological management of dementia puts a burden on those who are taking care of patients that suffer from this chronic condition. Caregivers frequently need to assist their patients with activities of daily living. However, they are also encouraged to promote functional independence. With the use of a discrete monitoring device, functional independence is increased among dementia patients while decreasing the stress put on caregivers. This paper describes a tool which improves the quality of treatment for dementia patients using mobile applications. Our application, iWander, runs on several Android based devices with GPS and communication capabilities. This allows for caregivers to cost effectively monitor their patients remotely. The data collected from the device is evaluated using Bayesian network techniques which estimate the probability of wandering behavior. Upon evaluation several courses of action can be taken based on the situation’s severity, dynamic settings and probability. These actions include issuing audible prompts to the patient, offering directions to navigate them home, sending notifications to the caregiver containing the location of the patient, establishing a line of communication between the patient-caregiver and performing a party call between the caregiver-patient and patient’s local 911. As patients use this monitoring system more, it will better learn and identify normal behavioral patterns which increases the accuracy of the Bayesian network for all patients. Normal behavior classifications are also used to alert the caregiver or help patients navigate home if they begin to wander while driving allowing for functional independence.

3.2 Introduction

Dementia is the loss of cognitive functioning. This effects the person’s daily life and activities such as the ability to think, remember and reason. Alzheimer’s disease, the most common form of dementia, is progressive and slowly destroys the brain limiting functionality, decreasing quality of life and eventually leading to death. [35] With advances in medicine and technology the number of older adults and life expectancy is increasing. The amount of people across the world with Alzheimer’s disease is predicted to almost double every twenty years [36]. Currently, an estimated 5.3 million Americans have Alzheimer’s disease. This is one out of eight people over the age of sixty five. In the year 2050, the number of Americans over eighty five will quadruple, leading to 959 thousand new cases of Alzheimer’s in that year alone [37]. Cost of caring for dementia patient is steep. While the cost increases with more severe stages of dementia the average lifetime cost of care for a single person is greater than $174,000. Total direct

15 and indirect costs in the US is over $100 billion annually. Businesses are spending $24.6 billion in health care while Medicare expenditures are an estimated $160 billion. Seven out of ten people with the disease live at home where 75% of costs are absorbed by the family [38]. In addition to incurring a large financial burden, the family also assumes responsibility as the primary caregiver, often an emotionally and stressful task. Caregiving may also have a negative impact on health, employment, income and financial security. About one third of family caregivers showed signs of depression, while half reported effects caused by caring to be their major health problem. Although the majority of caregivers live close to their patient, about 15% are considered long-distance caregivers, living over an hour away [37]. This makes constant monitoring of patients extremely difficult these remote caregivers. Current systems require the user to carry dedicated units which may be a GPS or push button alerts. They are monitored by a third party service whom charges a monthly fee. The GPS units are only used to locate the user or confine them to a specific radius, they do not issue dynamic alert automatically. Push button system are also limiting because they must be manually pressed by the user. The presented system, iWander, is designed to partially alleviate stress, financial burden, and offer easier remote monitoring to caregivers by using the user’s social network as a monitoring service. It is an application which runs on any Android enabled device that possesses the correct hardware. Most Android devices are commercial products, which amortized against total sales will keep the cost per unit down. The application runs in the background and collects data from the device’s sensors such as GPS, time of day, weather condition, stage of dementia, and user feedback. This data is then evaluated using Bayesian network techniques to determine the probability the person is wandering. Depending on the probability, iWander automatically takes actions that help navigate the patient to a safe location, notify caregivers, provide the current location of the patient and call 911.

3.3 Methods and Design 3.3.1 Software This application is designed on the Android software stack produced by Google. Android is an open source framework designed for mobile devices. It packages an operating system, middleware, and key programs [26]. The Android SDK provides libraries needed to interface with the hardware at a high level and make/deploy Android applications [27]. Application are written in Java and use SQL databases to store persistent data. We choose this platform as opposed to others because of the ability to easily thread background running processes, the polished Google Maps and Navigation API, and compatibility with other Android devices. Unlike dedicated systems this software is intended to integrate with the device’s existing applications; iWander must share resources with other applications. To make for a pleasant integration, it runs as inconspicuously as possible while using limited resources. We launch a background service that requests the GPS location and other data. Only when the probability of wandering is high will the activity wake up and interrupt the patient. Based on the probability evaluation and patient’s response the app can take diﬀerent actions. Which allows iWander to run harmoniously on the system while minimizing memory consumption and providing ease of use to the patient.

3.3.2 Wander Detection Wandering occurs because many dementia suﬀers have hypertension and feel a compelling urge to walk. Unfortunately, roughly 40% of them get lost when they do wander [39]. In order to identify a patient that is in danger of wandering the application creates safe zones, which are considered to be indoor or home locations where the patient is safe from the potential harms of wandering. These zones are identiﬁed by monitoring the GPS locations for areas where the phone is charged for

16 extended periods of time. The safe zone’s radius is relatively small and is only intended to enclose the patient’s dwelling. However, the exact size can be adjusted for extenuating circumstances. If the patient is inside a safe zone the app remains transparent. Once the patient is outside of the safe zones the probability of wandering is determined using Bayesian network techniques. A Bayesian network is a model for determining the probability that an event occurs given other variables of interest. As the variables change, inference can be applied to determine the likelihood of a specific event occurring [40]. The collection of relevant information allows predictions to be made with a larger degree of certainty. Overtime, more and more information is collected, and the occurrence of false positives is reduced [41]. The variables for the Bayesian network are chosen carefully because they directly effect the probability estimation. Researchers have shown the correlation between wandering and age, severity of dementia, and time of day [42]. iWander also adds time of day, current weather conditions, and time outside into this equation. Analyzing these variables against a collection of exsisting data can help classify behavior as normal or abnormal. There are several methods for classifying such cases. Currently we are choosing to use a two step approach described in [43], however our system is versatile enough to change methods during testing. First, the data is passed through a one-class support vector machine (SVM) [44, 45]. This filters out obvious, normal activity. It is important to pick proper sensitivity parameters to achieve a good trade off between false positive and false negative rates. To track successes and failures a simple prompt is issued ,after an alert, requesting feedback about correctness. The system can use this feedback along with usage statistics to dynamically adjust sensitivity. After the SVM, data is then passed to a nonlinear regression function which will finally classify behavior as normal or abnormal. The network shown in figure 1 displays the relations between the variables and how they affect the probability of wandering. Time of day greatly impacts wandering because about half of dementia patients experience night time walking of which a significant number result in the patient getting lost [16]. Therefore patients are more likely to wander at night. Dementia patients are outside less during night hours so length of time outdoors and weather also affect the probability of wandering. Lastly, dementia worsens with age [35]. Older adults have more severe dementia and suffers from the latter stages of dementia are more susceptible to wandering. Implementation of the Bayesian network is done in three phases. First is the pre-installation phase. This is the period of time where the network is static and the base case solely affects the probability of wandering. The relationships and statistics between the variables are the only inputs. Second is the configuration phase. Setting up the device for that specific patient is necessary as some information such as age and level of dementia is given as user input. The remaining information like safe zones, GPS, time of day and weather can be implicitly gathered by the device and geared of the individual patient. The last phase is the extended learning. Once the devices are deployed they continually learn from the gathered data. This data not is only tailored for the individual patient, but also has an affect on the baseline model. Thus the more the network is used the higher the accuracy is for all patients, an inherent property of Bayesian networks [46].

3.3.3 Alert Actions If the probability suggests the patient is likely wandering, action is taken. First, a notiﬁcation prompts them to provide feedback if they are okay. It has been shown that interrupts may help bring them out of a demented state [47]. The single option, “Yes, I am Okay,” is given. With positive feedback the alert process will pause, this also reduces false positives. If the patient does not respond, it is inferred that he or she is not okay and may be lost. The GPS coordinates are used, with Google Maps and Navigation tools, to give them directions to the safe zone. If no progress is made after a set period of time an alert is sent to the caregivers. Wanderers may not be properly prepared for harsh weather conditions such as extreme hot and cold temperatures and thus faster action must be taken [48]. The device can easily retrieve current weather conditions from reliable,

17 Figure 3.1: Bayesian Network Variable Relations Aﬀecting Wandering Probability

18 dedicated services via the Internet. This course of action is also taken if the time spent out of the safe zone is to great. The purpose of the alert is to notify the caregiver of wandering behavior, establish bidirectional communication with the patient and relay information. When an alert is issued the device calls a single Google Voice number. Google Voice will then ring several diﬀerent caregiver numbers simultaneously. When any of those caregivers answer they will be connected and the patient’s device will automatically be placed on speakerphone [49]. Enabling bidirectional communication between the patient and caregiver further reduces the number of false positives. Using Google Map APIs, longitude and latitude can be converted into a street address and relayed to the caregiver in several ways including SMS messages, email, or web interface updates. By communicating with the patient and evaluating location information the caregiver can plan an appropriate course of action. In urgent cases, the caregiver can give a command to instruct the patient’s device to 3-way a call to 911. By calling 911 from the patient’s device, it can be certain that the patient’s local 911 station will be called. This process allows caregivers to reduce the number non emergency calls to 911 and provide long distance, remote monitoring to the dementia patients.

3.3.4 Automotive Travel If the patient is traveling in an automobile it is highly probable they will travel outside of their safe zone. To limit the number of false positives the speed in which the patient is traveling is checked. This is done through the Haversine formula [50], which divides successive GPS coordinates by the time interval between them. Once the patient is determined to be in an automobile the Bayesian network is not used. Instead, iWander gathers specific information on that individual person. Many older people with dementia are not supposed to be driving, however they often do [51]. Driving is a sensitive issue that should be discussed between the patient and their primary caregivers [52]. iWander could send an alert every time the patient is driving however patients may reject the system if there are conflicts regarding driving privileges or they may simply be a passenger. Instead, iWander tries to prevent the patient from getting lost and offers location information to the caregivers which may promote functional independence. If the patient is determined to be an automobile two things are noted, time of day and common routes. By learning basic travel patterns alerts are only issued in abnormal situations such as traveling along an unfamiliar route at uncommon times. Issuing an onscreen notification which forces them to press a button may be distracting. Instead the app audibly prompts them offering directions if needed. Using Google Voice recognition and Google Navigation it searches for and gives them turn by turn directions to their destination.

3.4 Challenges 3.4.1 Hardware Designing on the Android platform provides several hardware related beneﬁts. New hardware with additional sensors, increased performance, and better battery life are constantly being released. The system works under the assumption that the patient carries the device on them. While the likelihood of remembering to carry a cell phone is higher due to common society standards, this may not be the case with all dementia patients. Often times phones are too bulky or forgotten [53]. A viable solution is to use an Android enabled watch. It possesses all the capabilities of an Android smartphone, but condensed into a watch version. Once attached, the patient is more likely to have it on them as many older adults are already accustomed to wearing watches for fall emergencies. An Android watch can automatically detect falls as well as wandering and can replace current push button watch alarms [54]. Due to the small screen size iWander would have to adjust screen displays to the user. Since the system is designed to reduce interactions with the user this change would be minimal.

19 3.4.2 Patient Interaction To overcome potential learning curves iWander operates discreetly and require minimal patient interaction. The majority of interaction is done through simple prompts, voice commands, and text to speech. Older adults often have diﬃculty with their haptic perception. To avoid the need for the software keyboard on the device, input is provided through the use of easy to press buttons and auditory input when applicable. By request of the patient the text on the screen can be read aloud through built in text to speech libraries. Lastly, all color schemes have high contrast ratios to promote ease of readability [55].

3.4.3 Implementation Accuracy of GPS is a major concern. However, previous dementia wandering systems using cellphone technology have proven the GPS accuracy to be reliable enough for similar systems [53]. The only remaining problem is a loss of connection. In the event of satellite connection is lost, the device can implement coarse-grain GPS which uses nearby antenna stations to triangulate the location of the device. This provides a less accurate alternative to the fine-grain, satellite GPS, while the connection is being reestablished [56]. If the fine-grain and coarse-grain GPS both fail, the device will broadcast the last known location when necessary. Last known location within minutes is usually acceptable for locating wandering persons because their position does not rapidly change [57]. Although battery technology is rapidly improving, the frequent use of GPS shortens the battery life. To prevent rapid battery loss the GPS is polled. For these purposes receiving new coordinates once every 5-10 minutes is sufficient [57]. When wandering is detected, the GPS locations can be obtained more frequently to monitor the patient closely. Another method to avoid low battery life is to remind the patient to charge if possible instead of waiting until the battery is critical to issue a reminder. If the patient is in a safe zone the battery can be charged and the user is prompted.

20 CHAPTER 4

BEAT: BIO-ENVIORMENTAL ANDROID TRACKING

4.1 Overview

We introduce BEAT (Bio-Environmental Android Tracking), which provides methods for collecting, processing, and archiving ones daily vital and spatiotemporal statistics using oﬀ-the-shelf wireless devices and biologic and environmental sensors. BEAT can operate in a self-contained manner on a mobile device and analyze vital information in real time. It uses statistics such as heartbeat variance and range thresholds to issue alerts. Alerts are propagated in a tiered fashion, so that the end user and his/her social contacts have a chance to detect false alerts before contacting medical professionals. BEAT is built on the open Android platform to support a diverse class of mobile devices. The framework can be extended to a fullﬂedged personal health monitoring system by incorporating additional biosensor data such as blood pressure, glucose, and weight.

4.2 Motivation

4.2.1 Heart Disease Heart disease is a leading cause of death in most developed nations. In the U.S., cardiac-related fatalities rank rst, being responsible for 26% of all deaths [58]. The most common sub-group of this category is coronary heart disease, attributable for 17.5% of all fatalities [58]. Coronary heart disease is the decreased or diminished ability to provide circulation to the cardiac muscle, which can lead to cardiac attacks and arrests. Each year, there are 1,255,000 coronary attacks, with 1/3 of them fatal [58]. However, 20% of mortalities can be prevented with personal monitoring systems [59]. Traditional personal monitoring systems typically use laptop computers and/or external storage media later connected to a desktop computer for analysis. Newer personal monitoring systems incorporate smaller mobile computing units (e.g., cell phones). However, many systems are not interactive [60]; gathered sensor information lacks real-time feedback from the monitored users. Others tend to rely on third-party service for data storage and analyses, which impose unnecessary constraints on the deployment model and associated power consumption. The advent of smartphones, with their continual increase in computing power, storage capaci- ties, and builtin sensors, oﬀers a unique opportunity to address the above constraints. We introduce BEAT (BioEnvironmental Android Tracking), which exploits the increasing capabilities of smartphones to support local storage and analyses of data, as well as real-time monitoring, feedback, and emergency response.

21 4.3 Current Systems

Several systems have been developed to bring medicalgrade data gathering and analysis out of the hospital and into the home. The system presented in [61] incorporates mobile devices and uses threshold-based algorithms to detect life-threatening arrhythmias. Others have used sensors to monitor medical conditions such as obstructive sleep apnea [62]. Even detection of stress levels [63] has been proposed using threshold-based personal monitoring.

4.3.1 Wireless Body Area Networks Wireless body-area networks are intended to be composed of passive sensor nodes, active actuator nodes, and a wireless personal device [64]. The passive nodes collect data (accelerometer, heart rate, etc.), and the active nodes perform operations (dispense medicine, release insulin, etc.). Both communicate with a personal device for more complex operations. However, the focus has been on system architecture and service platforms for extra-body communication [65]. Larger challenges are how to engage patients in a dialog about their health, and how to make it easy for patients to manage their chronic care.

4.3.2 m-Health The MobiHealth system has developed a way to transmit vital signals over public wireless networks to health-care providers. It relies mostly on an m-health platform that connects medical professionals with end users [60]. The use of third-party monitoring services may be problematic due to the lack of bandwidth and coverage in the current network. The high power consumption required for transmitting large data ﬁles wirelessly also highlights the need for processing data on the mobile device whenever possible.

4.3.3 Fitness The ﬁtness industry has also adopted personal monitoring systems that track weight and physical activity levels. Other devices have been created that use accelerometers as pedometers and as well as various wristwatch-like devices to display heart rate. Several of these systems operate with an external storage device, allowing additional processing, user inputs, and other communications.

4.4 BEAT Design

BEAT provides methods for collecting, processing, and archiving ones daily vital and spatiotemporal statistics by integrating an Android device with commercially available biologic and environmental sensors. With all the capabilities of modern smartphones, data can be processed and analyzed locally, to alleviate network bandwidth overhead [66]. This approach not only saves power, but also allows real-time interactions with the user to further tune system behaviors.

4.4.1 The choice of mobile operating system The choice of operating system (OS) is critical when designing an open framework that will be deployed on multiple device types. While Symbian and RIM devices continue to dominate the global smartphone market [67], their market shares, along with Windows Mobile, have experienced declines, while Android and iPhone are gaining rapidly. Android and iPhone have sophisticated application marketplaces, which provide easy access to both software distribution and maintenance channels. Though the iPhone platform is more mature, Android has been gaining considerable momentum, with 70,000 applications and over 1 billion downloads as of July 2010 [68].

22 There is a major distinction between the two promising OS options. In contrast to Apples proprietary source code, Googles Android is open-source, which provides developers access to OS code. This paradigm enables rapid development of prototypes that can utilize system resources. Programming the device at both the application and OS levels provides superior customization and functionality over the iPhone OS. Android also utilizes the Java programming language at the application layer, which enables applications to be run in a Java Virtual Machine. This allows applications to behave consistently on every device, reducing platform dependencies and backwards compatibility problems [27].

4.4.2 Real-time monitoring BEAT performs analyses of both short-term and long-term patterns and provides notications to the user. Threshold-based algorithms are used for real-time monitoring. If the beats per minute detected deviate from user-dened thresholds, the unit invokes a pre-dened handler function to raise an alarm. For example, if the user is exercising and their heart rate exceeds a desired range, BEAT will notify the user to slow down. The data is also stored at user-specied granularities for inspection and trend analysis. Optionally, the data can be uploaded to health professionals.

4.4.3 Emergency response In the event of an emergency, notications are rst issued to the user, who may identify a false positive (e.g., monitoring strap removal, adjustment). If there is no response within a dened time, communication to the users pre-specied social contacts is attempted. If contact is made, social contacts can evaluate the situation further and alert emergency personnel. By using this multilevel- ow, emergency alerts can be canceled by the user. All threshold values can be customized based on individual needs [49].

4.4.4 Contextual data analyses and user interactions BEAT categorizes events by processing the collected data. Based on the current state and patterns of the biologic and environmental sensors, BEAT can prompt users to enter additional data. To illustrate, if the heart rate monitor detects an increase in pulse rate, the accelerometer detects movement, and the GPS detects a frequent change in location, it can be inferred that the user is exercising. Threshold parameters are then adjusted automatically to minimize interactions with the user. In cases where only heart rate may spike, with no accelerometer and GPS activity present, the user is prompted to provide simple feedback on his/her physical symptoms and whether BEAT has raised a false alarm. This feedback is stored to be used in machine-learning algorithms for guiding future interactions, to reduce false alarms and unnecessary interactions [54]. This design is comparable to similar emerging systems such as automated wandering detection for dementia patients [69]. For such cases, the data set can be agged for further analysis by medical professionals

4.4.5 Long-term analysis Long-term patterns can be extracted from incremental base-line data checkpoints over time. Even gradual changes can be tracked, thus enabling the earliest possible detection of an emerging condition. With the users permission, health history and analysis can be uploaded from the device to health professionals periodically or on demand.

4.4.6 Power management Another requirement of the BEAT framework is low power consumption. This relates to both the collection of external sensor data as well as the transmission of this data to third parties, as

23 needed. Usertunable parameters can also help reduce the polling frequency if verbose data is not required. Uploading sensor data to medical-care providers also inuences the power overhead of the system. Data compression can help limit the size of the required transmission, thereby reducing power to transmit. Unfortunately, the compression process itself requires power to perform. To sidestep this constraint, compression, analysis, and transmission can be deferred until the mobile device is connected to an external power.

4.5 Implementation 4.5.1 Hardware The BEAT framework uses the Zephyr Bluetooth HxM monitor - a small, lightweight sensor, with embedded power supply worn across the chest. It utilizes the conductive Zephyr Smart Fabric strap and embedded ECG electronics to calculate heart rate. The HxM monitor provides 24 hours use per battery charge, with a Bluetooth operating radius of 30 ft.. In addition to ECG electronics, the monitor is equipped with a 3-axis solid-state accelerometer, which can additionally calculate strides, distance covered, and speed data. The monitor operates in Bluetooth slave mode. Once a connection is established with an Android device, the monitor will continually broadcast transmissions at one second intervals through the duration of the connection.

4.5.2 Software BEAT was developed using the Android Framework API level 6, targeting device platform versions 2.0 and newer. The Linux kernel versions varied between devices tested, ranging from 2.6.29 to 2.6.35. Initial prototypes of the BEAT system required customized kernel builds; however, with the Android 2.0/2.1 release (API level 5) this requirement was removed. The release provided BEAT native access to the Java Bluetooth API, enabling enhanced functionality such as service discovery and the ability to create the necessary Bluetooth serial (RFCOMM) channel with the Zephyr device. Removal of the custom kernel requirement allows the BEAT system to be deployed on any existing Android device 2.0 and newer. The BEAT framework consists of a set of interoperating applications including a system service, a data viewer, a data logger, and emergency alert (Figure 1). Communication among the separate entities of BEAT is achieved through the use of a shared data channel called a ContentProvider and through Intents, which are messaging structures used to launch Activities (user interactions), initiate services, or broadcast communications. Modular BEAT components register BroadcastReceivers that respond when specific Intents occur. The ContentProvider contains an embedded SQLite database and provides a common interface to all Android packages. At system initialization, BEAT first connects the phone with the heart-monitoring device. A dedicated user Activity allows the phone to scan, pair, and connect with the device using the Android Bluetooth API. The Bluez kernel sub-system utilizes the hciattach daemon to connect to the hardware-specific UART driver. Once paired with an Android device, a RFCOMM channel is established, and prior to Activity termination, a dedicated thread is spawned to maintain this communication channel. When the thread detects an incoming packet, it first performs a CRC check to ensure data integrity. Data packets that fail the check are discarded. Acceptable packets are parsed and packaged into an Intent object. The Intent is assigned a HEART BEAT action and is broadcast to the system. Multiple components of BEAT register BroadcastReceivers responding to this Intent, including the data viewer and logger. On receipt of a HEART BEAT Intent, the live data viewer application updates its values to reflect the latest received heart-rate data. The gathered data can be viewed both numerically and in time- series graphs within this interface. The BEAT logging component also contains a BroadcastReceiver

24 Table 4.1: Projected number of data points for each sensor

Heart Rate GPS Accelerometer Minute 3 2 180 Hour 180 120 10,800 Day 4,320 2,880 259,200 Week 30,240 20,160 1,814,400 Month 907,200 604,800 54,432,00 Year 10,886,400 7,257,600 653,184,000 registered to listen for the same HEART BEAT Intent. Once received, the logger stores the data from the Intent in an internal buffer. A user configurable parameter (default 60 sec/write) determines the write frequency from this temporary storage into the BEAT ContentProvider and underlying SQLite database. If BEAT identifies that heart-rate exceeds predetermined thresholds, it initiates the Alert process by broadcasting an Intent of ALERT action. This Intent starts a dedicated Alert Activity that waits for user feedback for dismissal of a false positive. This time window was selected to be 10 seconds (configurable). If response is not made within this interval, the alert Activity retrieves emergency contact information from the BEAT ContentProvider and issues a CALL Intent and enables speakerphone via the systems AudioManager.

4.6 Evaluation

Evaluation of the BEAT framework was performed by measurement of the storage consumption rate of logging data from multiple sensors, and power consumption due to local data analyses and transmissions.

4.6.1 Storage While short-term monitoring and archival of data may require a trivial amount of storage space, longterm storage necessitates additional design considerations. The data storage requirements (Ta- bles I and II) are well within the range of todays devices, even without employing ltering techniques or data compression. LempelZivWelch compression was used, however, to reduce the amount of data required to be transferred.

4.6.2 Power Overhead Power consumption of the BEAT framework was measured using PowerTutor [70]. Also tracked were the Android Dalvik log and active applications. This additional data ensured that the power overhead accurately depicted the incremental draw of the BEAT framework and Bluetooth sub- system. The Android devices used for power proling included the HTC G1, Nexus One, Droid Incredible, and EVO 4G. At present, the BEAT system requires 550 mW for operation. This is largely due to the Bluetooth sub-system and remains a major issue for the BEAT system. However, with the upcoming release of Android 3.0, we expect improved power eﬃciency.

25 Figure 4.1: BEAT Component Overview

Table 4.2: Size estimation of data from various sensors (KB)

Day Week Month Year Raw 22 154 4,620 55,440 Heart Rate Compressed 6 42 1,260 15,120 Raw 6 42 1,260 15,120 GPS Compressed 3 21 630 7,560 Raw 310 2,170 65,100 781,200 Accelerometer Compressed 150 1,050 31,500 378,000

26 CHAPTER 5

CONCLUSION

The number of older adults has been setting record highs. Unfortunately with age comes the likelihood of chronic conditions. Today, there exists a number of technologies that address these problems. However, these technologies are often lacking in functionality, extendability, and are attached to high price tags. Current smartphone devices are fully capable of oﬀering viable oﬀ the shelf solutions. They can be loaded with applications that assist with these problems.

5.1 Applications 5.1.1 iFall Falling is the most common problem older adults face. This application provides a viable solution to fall detection. Tri-axial accelerometers embedded in the device relay information about the gravitational forces. By performing the root-sum-of-square the total amount of force can be calculated. This total is noted and matched against the know signature of a fall. Once this pattern is detected, an actual fall is likely, and the alert process is activated.

5.1.2 iWander This application is for dementia patients that have wandering problems. It uses the geo- positioning satellites along with contextual information. This contextual information includes things like time of day, weather, and distance from safe zones. The data is passed into a learning network and correlated to classify behavior as normal or abnormal. A estimated probability of wandering is determined and automatic actions are taken. The actions range from attempting to self correct the user to alerting caregivers.

5.1.3 BEAT By leveraging the power of Android, the BEAT system provides the short-term information needed for users and caregivers to react in real time to, or even prevent, life threatening events. It also provides the long-term information needed for health-care providers to make lifestyle and medication recommendations. In combination, these will aﬀord the user a more independent and higher quality of life. The BEAT system can be tailored to the individual, and with the ability to integrate additional sensors, the system has the potential to do even more.

5.1.4 Alert For all intensive purposed the alert sequence can be an independent application. It simply waits until it receives a signal to start the process. Once this signal is triggered it attempts to gain the users attention by ringing, vibrating, and ﬂashing the screen and on-board LEDs. The user is then

27 presented with the single option to cancel the alert. This acts as a simple mechanism to give the end user the initial chance to reduce false positives. If they do not respond the next step in the process if triggered. Automatically notifying loved ones the second step in the alert process. Instead of calling all the loved ones (simply just pre-specified social contacts) one by one, advanced capabilities of Google Voice are utilized. Google Voice offers free phone numbers and call forwarding. This free number is designated as the emergency number the alert app will call if needed. Once configured properly, Google Voice will then simultaneously forward that call to several contacts at once. By allowing So- cial Monitoring the need for dedicated third party monitoring services are avoided, thus minimizing the cost of the system.

5.2 Current State 5.2.1 Human Subject Testing In corporation with Dr. Ken Brummel-Smith, Chair of Florida State University’s Department of Geriatrics [71], and the FSU Wrestling Club we have performed human tests. These tests were designed to collect accelerometer data during a real fall. A real fall is the closest we are able to safely reproduce. We are assured that the members of the wrestling club are trained in the art of falling safely. A helper application was created to log accelerometer points to the SD-card while the subjects fell to the ground in various ways. The test was perform with three diﬀerent Android Devices: the Motorola Droid-2 , the Motorola Droid-X, and the Samsung Captivate. The devices were put in the subject’s right pocket and attached to their right forearm and ankle using ace bandages. They were then instructed to fall forward, backwards, and side to side on a standard wrestling mat. The falls were video recorded to allow for visual reference. We then rotated the devices positions and repeated the procedure. These tests were completed with two individuals of diﬀerent body types. The data is currently being evaluated and the information is scheduled for upcoming release.

5.2.2 Real World Evaluation We are currently working with Westminster Oaks, an assisted living facility, in Tallahassee, Florida, to deploy devices running our systems. Focus groups are planned to discuss special interface and procedural needs. The alert protocol will be modified slightly since the on-staff nurses are the primary caregivers and dialing 911 is not needed. Our goal is to mine enough data to create an efficient baseline model to detect problems with good accuracy. Once openly released to everyone, the baseline model will be modified based on learning to increase accuracy.

5.2.3 APIs, Extendability, and Licensing Since publicizing iFall we have received many inquiries from vendors about integrating iFall into their current products and technologies. Due to this demand an API has been created that allows third party application developers to remotely start and stop the iFall service. While running, the iFall service provides broadcasts about what state the user is currently in. This states include:

1. Base-State: The user is standing straight up.

2. Falling-State: The user is in free fall to the ground.

3. Impact-State: The user has come in heavy impact with the ground.

4. LongLie-State: The user has not recovered from impact and is on the ground

5. Alert State: The user has been in the LongLie State for an extend period of time.

28 These APIs can be found at :

ww2.cs.fsu.edu/ sposaro/iFall There has been number of other areas that can beneﬁt from iFall technology. Aside from the elderly, we have been contacted to integrate and extend iFall for other applications. These applications include horseback riding, epilepsy, mountain biking, and even an adult tricycle company. We currently oﬀer developers the ability to use the basic iFall monitoring service free of charge. It is openly available on the Android Marketplace. Currently, we are taking requests and feedback to further develop this application.

5.3 Future Works 5.3.1 Additional Accelerometer Sensors The ﬂexibility of the Android platform along with the phone’s hardware capability allows this system to be extended in numerous ways. Bluetooth support could allow iFall to gather additional data readings from micro-sensors embedded in articles of clothing [32]. Ideally, a sensor would be embedded in head or eye wear due to the fact that the head is the most reasonable location for fall detection using threshold based algorithms [34].

5.3.2 Activity Classiﬁcation More sophisticated pattern matching algorithms can be ran [33]. Eﬀorts are being made to build a database of common ADL readings [23]. This information can be exploited in attempts to classify what type of action the user is performing based on pattern matching techniques. This actions include everyday activities such as walking, sitting, and driving. The system could also use image support from mounted, Bluetooth cameras as described in [17], [18], and [19].

5.3.3 Daily Medical Monitoring BEAT is also designed to be extensible. As the popularity of mobile devices grows, so does the number of various sensors on the market. Additional sensors can be incorporated to gather other vital readings such as blood pressure, glucose, and weight [72]. Bio-sensors capable of monitoring these variables can be paired via Bluetooth radio and integrated into the BEAT framework.

5.3.4 Depression Many patients with dementia are at a higher risk for depression. We can monitor patient behavior such as movements (using Bluetooth technology [73]), communication patterns and level of activity. Once a normal pattern is established we can use a learning algorithm to detect non-normal activity. For example: the patient is normally moderately active, makes several trips to the kitchen, dials and answers calls to family members an average of three times a day, and registers high accelerometer readings. If they stop moving, spend a lot of time in the bed room, stop eating and start ignoring phone calls, that abnormal behavior may be a sign of depression. The device is able to issue a version of the standard depression index test to mine feedback from the user and further evaluate the situation.

5.3.5 Server Data Analysis The large volume of uniform data collected by the system from multiple individuals can also be of importance to medical professionals It can WiFi connection to log the data readings from several users on a server. This enables the ability to perform correlation studies on observed variables.

29 New early warning signs for life-threatening conditions may be identied based on these studies. In addition, there is the potential for more timely and detailed feedback for large scale medical studies of new treatments or medications.

5.3.6 Vision Aid Not only can modern Android devices be used for passive monitoring, they can have a highly active role. Using the camera, speaker, and on-board processing power Android devices can help the visually impaired see the world like never before. By taking photos from the camera and converting them to black and white pixels, algorithms can then be ran to turn these pixels into some form of audio feedback. The lighter the pixel the higher pitch the noise. Conversely, the darker the pixel the lower pitch the noise. After some training, this enables the blind to experience sights as they were previously unable to do. [74]

30 APPENDIX A

DOCUMENTS

31 A.1 Human Subjects Testing Approval

32 A.1.1 IRB Continued

33 A.2 Human Subjects Testing Renewal

34 A.3 Support for Testing

35 A.4 Testing Consent Form

INFORMED CONSENT FORM

A Study on Detecting Falls Using Cell Phones

Dear Participant:

My name is Gary Tyson, PhD and I am a faculty member in the Department of Computer Sciences at Florida State University. I am conducting a study about how movement sensor technology in cell phones may be able to help detect a person falling. If cell phones can accurately detect falls, then the “iFall” cell phone could be used to help older adults who have fallen by notifying emergency contacts. This project involves collecting data measurements of individuals’ falling while wearing the iFall phone in various locations on their person. Each participant will be asked to fall onto wrestling mats a minimum of 6 times wearing the phone in a front pants pocket, wearing the phone on a belt pouch, and wearing a phone placed on the forearm and calf using an ace bandage. All falls will be video recorded.

The phone will automatically record data during the fall then the data will be later integrated into a data set containing accelerometer readings time synced with videotape of the falls. The data set will then be used to evaluate various fall detection heuristics, as well as to determine differences in reading due to monitor placement at different body locations. The data set will be made available to other fall detection researchers nationally. This study will be conducted on Florida State University’s campus and take approximately 30 minutes. Instruction will be provided by me and graduate students associated with the project.

You have been invited to participate because you have confirmed that you are at least 18 years old and participate in the FSU wrestling club. All study procedures do not involve activities that would cause additional risk or discomfort to you that you would not otherwise be exposed to in a regular practice session. The iFalls phone will be attached to your person, but padded for your protection. If you do experience any injury while taking part in the study you will have access to medical services provided by the FSU wrestling club trainers. A trainer will be on-site when you are participating in the study. The researchers involved in this study do not assume financial liability for your injuries. Should a cell phone be damaged during a fall you will not be held liable for any damages. If you choose to participate, you have the option to stop at any time without any penalty or risk. A donation in the amount of $200 will be given to the FSU Wrestling Club by the researchers regardless of your decision to participate.

All forms and digital recordings containing your responses for the study will be kept in a locked filing cabinet in my office at Florida State University. Only I, the staff, and the graduate students working on this study will have access to them. To maintain confidentiality of your records, I or one of my staff or graduate students will assign an experimental code to your paperwork. The results of this research study may be published but your name or identity will not be revealed. Only group findings will be reported. Confidentiality will be maintained to the extent allowed by law.

There are several potential benefits from participating in this research. Data gathered may be used to help improve medical monitoring capabilities for falls detection. Early detection of a fall event will greatly increase the possibility of an improved outcome. This study will advance the development and shorten the time frame for developing accurate fall sensors.

FSU Human Subjects Committee Approved on 12/09/10. Void after 12/07/11. HSC# 2010.5431

36 A.4.1 Consent Continued

37 BIBLIOGRAPHY

[1] Alan Greenblatt. Aging baby boomers. CQResearcher, 17(37), October 2007.

[2] M.P.H. Arialdi M. Minio. Death in the united states, 2009. NCHS Data Brief, 64, July 2011.

[3] Alliance for Aging Research. The silver book: Osteoporosis. Chronic Disease and Medical Innovation in an Aging Nation.

[4] Alliance for Aging Research. The silver book: Parkinsons disease. Chronic Disease and Medical Innovation in an Aging Nation.

[5] Eric A. Taub. Helping grandpa get his tech on. NY Times. Person Tech, October 2009. http://www.nytimes.com/2009/10/29/technology/personaltech/29basics.html.

[6] Life Alert Emergency Response. Consumer aﬀairs. www.consumeraﬀairs.com/age/lifealert.html, November 2009.

[7] Matrix Interactive LLC. Lifelink. www.callforassistance.com, November 2009.

[8] Department of Health and Human Services. Fatalities and injuries from falls among older adults - united states, 1993-2003 and 2001-2005. pages 1221–1224, November 2006. Morbidity and Mortality Weekly Report.

[9] Garret Brown. An accelerometer based fall detector: Development, experimentation, and analysis. Technical report, July 2005. EECS/SUPERB.

[10] K.E. Thomas, J.A. Stevens, K. Sarmiento, and M.M. Ward. Fall-related traumatic brain injury deaths and hospitalizations among older adults - united state, 2005. Journal of Saftey Research, (39):269–272, May 2008.

[11] A.K. Bourke, J.V. O’Brien, and G.M. Lyons. Evaluation of a threshold-based tri-axial accelerometer. (26):194–199, September 2006.

[12] EW Peterson, CC Cho, and ML Finlayson. Fear of falling and associated activity curtailment among middle aged and older adults with multiple sclerosis. (13):1168–1175, 2007. Multiple Sclerosis.

[13] K Doughty, R Lewis, and A McIntosh. The design of a practical and reliable fall detector for community and institutional telecare. Journal of Telemedicine and Telecare, 6(1):150–154, 2000.

[14] Majd Alwan, Prabhu Jude Rajendran, Steve Kell, David Mack, Siddharth Dalal, Matt Wolfe, and Robin Felder. A smart and passive ﬂoor-vibration based fall detector for elderly.

[15] Mihail Popescu, Yun Li, Marjorie Skubic, and Marilyn Rantz. An acoustic fall detector system that uses sound height information to reduce the false alarm rate. 30th Annual International IEEE EMBS Conference, August 2008.

38 [16] Tracy Lee and Alex Mihailidis. An intelligent emergency response system: preliminary development and testing of automated fall detection. Journal of Telemedicine and Telecare, 11(4):194– 198, 2005. [17] Shaou-Gang Miaou, Pei-Hsu Sung, and Chia-Yuan Huang. A customized human fall detection system using omni-camera images and personal information. pages 39–41. Proceedings of the 1st Distributed Diagnosis and Home Healthcare (D2H2) Conference, April 2006. [18] Caroline Rougier and Jean Meunier. Demo: Fall detection using 3d head trajectory extracted from a single camera video sequence. [19] Hammadi Nait-Charif and Stephen J. McKenna. Activity summarisation and fall detection in a supportive home environment. 2004. [20] G Williams, K Doughty, K Cameron, and D.A. Bradley. A smart fall and activity monitor for telecare applications. volume 30, pages 1151–1154. Proceedings of the 20th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 1998. [21] Thomas Riisgaard Hansen, J. Mikael Eklund, Jonthan Sprinkle, Ruzena Bajcsy, and Shankar Sastry. Using smart sensors and a camera phone to detect and verify the fall of elderly persons. European Medicine, Biology and Engineering Conference (EMBEC 2005), November 2005. [22] N Noury, A Fleury, P Rumeau, A.K. Bourke, G.óLaighin, V Rialle, and J.E. Lundy. Fall detection - principles and methods. pages 1663–1666. Proceedings of the 29th Annual International Conference of the IEEE EMBS, August 2007. [23] G Pérolle, P Fraisse, M Mavros, and I Etxeberria. Automatic fall detection and activity monitoring for elderly. Sponsored by the CRAFT project. [24] Ran Wei and Ven-Hwei Lo. Staying connected while on the move. 8(1):53–72, 2006. [25] HTC Corporation. G1 product specification, 2009. [26] Google Inc. Android. www.android.com. [27] Google Inc. Android developers. developer.android.com. [28] Jay Chen, Karric Kwong, Dennis Chang, Jerry Luk, and Ruzena Bajcsy. Wearable sensors for reliable fall detection. pages 3551–3554. Proceedings of the 2005 IEEE Engineering in Medicine and Biology 27th Annual Conference, September 2005. [29] A.K. Bourke and G.M. Lyons. A threshold-based fall-detection algorithm using abi-axial gyroscope sensor. (30):84–90, December 2008. [30] Suhuai Luo and Qingmao Hu. A dynamic motion pattern analysis approach to fall detection. 2004. [31] J.Y. Hwang, J.M. Kang, Y.W. Jang, and H.C. Kim. Development of novel algorithm and real- time monitoring ambulatory system using bluetooth module for fall detection in the elderly. pages 2204–2207, September 2004. [32] M.N. Nyan, Francis E.H. Tay, M. Manimaran, and K.H.W. Seah. Garment-based detection of falls and activities of daily living using 3-axis mems accelerometer. Journal of Physics, pages 1059–1067, 2006. [33] Tong Zhang, Jue Wang, Ping Liu, and Jing Hou. Fall detection by embedding an accelerometer in cellphone and using kfd algorithm. IJCSNS International Journal of Computer Science and Network Security, 6(10):277–284, October 2006.

39 [34] Maarit Kangas, Antti Konttila, Ilkka Winblad, and Timo Jamsa. Determination of simple thresholds for acceleroetry-based parameters for fall detection. pages 1367–1370. Proceedings of the 29th Annual International Conference of the IEEE EMBS, August 2007.

[35] Alzheimer’s Disease Education and Referral Center. Alzheimer’s disease fact sheet. NIH Pub- lication No. 08-6423, February 2010.

[36] Times Online. Number of dementia suﬀerers will double every 20 years. October 2009.

[37] Alzheimer’s Association. Alzheimer’s disease facts and ﬁgures. Alzheimer’s and Dementia, 6, March 2010.

[38] Christine Kennard. Statistics about the ﬁnancial costs of alzheimer’s disease. About.com Health’s Disease and Condition, September 2006.

[39] Allegra Stratton. Charity backs tagging for dementia suﬀerers. December 2007.

[40] David Heckerman. A tutorial on learning with bayesian networks. Microsoft Research, March 1995.

[41] Los Alamos Center for Bayesian Methods in Environment, Safety, and Health. What is bayesian inference? September 1998.

[42] Diane A. Klein, Martin Steinberg, Elizabeth Galik, Cynthia Steele, Jeannine-Marie Sheppard, Andrew Warren, Adam Rosenblatt, and Constantine G. Lyketsos. Wandering behaviour in community-residing persons with dementia. International Journal of Geriatric Psychiatry, pages 272–279, April 1999.

[43] Jie Yin, Qiang Yang, and Jeﬀrey Junfeng Pan. Sensor-based abnormal human-activity detection. IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, March 2007.

[44] B. Scholkopf, J. Platt, J. Shawe-Taylor, and A. Smola. Estimating the support of a high- dimensional distribution. Neural Computation, 13(7):1443–1471, July 2001.

[45] D.M.J. Tax and R.P.W. Duin. Support vector domain description. Pattern Recognition Letters, 20(1113):11911199, 1999.

[46] S. Russell and P. Norvig. Artiﬁcial Intelligence: A Modern Approach. Prentice Hall, 1995.

[47] Katherine Heard and T. Steuart Watson. Reducing wandering by persons with dementia using diﬀerential reinforcement. Journal of Applied Behavior Analysis, 32(3):381–384, 1999.

[48] Myra A. Aud. Dangerous wandering: Elopments of older adults with dementia from long-term care facilities. American Journal of Alzheimer’s Disease and Other Dementias, 19(6):361–368, 2004.

[49] Frank Sposaro and Gary Tyson. Geriatric medical application suite on a sweet phone. First AMA IEEE Medical Technology Conference on Individualized Healthcare, March 2010.

[50] C.C. Robusto. The cosine-haversine formula. The American Mathematical Monthly, 64(1):38– 40, January 1957.

[51] G.K. Fox, F. Withaar, and G.M. Bashford. Dementia and driving: a survey clinical practice in aged care assessment teams. Australian Journal on Ageing, 15(3), 1996.

[52] Gary J. Kennedy MD. Advanced age, dementia, and driving: Guidance for the patient, family and physician. Primary Psychiatry, 16(9):19–23, September 2009.

40 [53] Oxford University Press on behalf of the British Geriatrics Society. Electronic tracking of patients with dementia and wandering using mobile phone technology. Age and Ageing, 34:497– 518, 2005.

[54] Frank Sposaro and Gary Tyson. ifall: An android application for fall monitoring and response. 31st Annual International Conference of the IEEE Engineering in Medicine and Biology Society, 1:6119–22, 2009.

[55] Arthur D. Fisk, Wendy A. Rogers, Neil Charness, Sara J. Czaja, and Joseph Sharit. Chapter 5: Design of input and output devices. Designing for Older Adults: Principles and Creative Human Factors Approaches, pages 65–72.

[56] Noam Shoval, Gail K. Auslander, Tim Freytag, Ruth Landau, Frank Oswald, Ulrich Seidl, Hans- Werner Wahl, Shirli Werner, and Jeremia Heinik. The use of advanced tracking technologies for the analysis of mobility in alzheimer’s disease and related cognitive diseases. BCM Geriatrics, 8(7), March 2008.

[57] Koichi Shimizu, Kuniaki Kawamura, and Katsuyuki Yamamoto. Location system for dementia wandering. Proceedings of the 22nd Annual EMBS International Conference, pages 1556–1559, July 2000.

[58] Murphy SL Xu J, Kochanek KD and Tejada-Vera B. Deaths: Final data for 2007. National Vital Statistics Report, 58(19), May 2010.

[59] Ruggerio C. Shah NB, Der E, Heidenreich PA., and Massie BM. Prevention of hospitalizations for heart failure with an interactive home monitoring program. 69th Scientic Session of the American Heart Association, 1997.

[60] Van Halteren A, Bults R, Wac K, Konstantas D, Widya I, Dokovsky N, Koprinkov G, Jones V, and Herzog R. Mobile patient monitoring: The mobihealth system. The Journal on Information Technology in Healthcare, 2(5):365–373, 2004.

[61] Gay V Ledijdekkers P. Personal heart monitoring system using smart phones to detect life threatening arrhythmias. 19th IEEE Symposium on Computer-Based Medical System, June 2006.

[62] Flores-Mangas F Oliver N. Healthgear: A real-time wearable system for monitoring and analyzing physiological signals. International Workshop on Wearable and Implantable Body Sensor Networks, pages 64–68, April 2006.

[63] Du CQ Picard RW. Monitoring stress and heart health with a phone and wearable computer. Motorola Oﬀspring Journal, 1:14–22, 2002.

[64] Barnard M van Dam K, Pitchers S. Body area networks: Towards a wearable future. Proc of WWRF kick oﬀ meeting, March 2001.

[65] Latre B., Braem B., Moerman I., Blondia C., and Demesster P. A survey on wireless body area networks. 2007.

[66] Leijdekkers P Jones V, Gay V. Body sensor networks for mobile health monitoring: Experience in europe and australia. The Fourth International Conference on Digital Society, February 2010.

[67] Miller M. Google android smacks down windows mobile in latest gartner data. May 2010.

[68] Zeman E. Android market: 70,000 apps, 1 billion downloads. InformationWeek, July 2010.

41 [69] Tyson G. Sposaro F., Danielson J. ifall: An android application for fall monitoring and response. 32nd Annual International Conference of the IEEE Engineering in Medicine and Biology Soci- ety, September 2010. [70] Qian Z. Zhang L., Tiwana B., Wang Z., Dick R., Mao ZM., and Yang L. Accurate online power estimation and automatic battery behavior based power model generation for smartphones. Proceedings of CODES+ISSS, October 2010. [71] Ken Brummel Smith. Charlotte edwards maguire, m.d., chair, and professor, department of geriatrics. The Florida State University College of Medicine. [72] Lau CT Agarwal S. Remote health monitoring using mobile phones and web services. Telemedicine and e-Health, 16(5), June 2010. [73] Ben Fagin, Frank Sposaro, and Gary Tyson. Indoor tracking of geriatric patients using bluetooth. First AMA IEEE Medical Technology Conference on Individualized Healthcare, March 2010. [74] Peter Meijer. The voice. www.seeingwithsound.com.

42 BIOGRAPHICAL SKETCH

I was born in September 13, 1986 in Philadelphia, PA. Being raised in South Jersey and South Florida I gained a great interest in computers and video games. I still remember being the only kid in my class that typed their vocab sentences using a Commodore 64. Mr. Joe Tosh first introduced me to programming freshmen year at Atlantic City High School. I took C++ plus over A+ because at the time figured C++ was a more advanced class because of the extra plus sign. After graduating from Bloomingdale High School in 2004 with honors I starting seeking a computer science degree at Florida State University. Creating video games was my initial motivation. Since than I have received a Bachelor’s Degree in 2008 and have found great interest in mobile programming. I am the initial student to start the mobile lab with Dr. Tyson. After working on the my first project, iFall, Dr. Tyson and I designed the Mobile Programming course as FSU. The course is used as a training base to recruit new students into the lab. I also went on to implement the redesign of the favorite contacts for Androids Ice Cream Sandwich at Google HQ in Mountain View, California. I am currently acts as a tech lead in the lab getting infrastructure and project management tools setup. I enjoy native Android coding and UI design.