MONITORING ARM USE IN DAILY LIFE VIA RFID TRACKING OF HOUSEHOLD OBJECTS JOYDIP BARMAN

Transcription

1 MONITORING ARM USE IN DAILY LIFE VIA RFID TRACKING OF HOUSEHOLD OBJECTS by JOYDIP BARMAN DALE S. FELDMAN, COMMITTEE CHAIR GITENDRA USWATTE, CO-CHAIR EDWARD TAUB ALAN W. EBERHARDT ALAN M. SHIH A DISSERTATION Submitted to the graduate faculty of The University of Alabama at Birmingham, in partial fulfillment of the requirements for the degree of Doctor of Philosophy BIRMINGHAM, ALABAMA 2012

2 Copyright by JOYDIP BARMAN 2012 ii

3 MONITORING ARM USE IN DAILY LIFE VIA RFID TRACKING OF HOUSEHOLD OBJECTS JOYDIP BARMAN DOCTOR OF PHILOSOPHY IN BIOMEDICAL ENGINEERING ABSTRACT After stroke, capacity to complete tasks in the treatment setting with the moreaffected arm is an unreliable index of actual use of that extremity in daily life. Available objective methods for monitoring real-world arm use rely on placing movement sensors on patients. These methods provide information on amount but not type of arm activity, e.g., functional vs. nonfunctional movement. This paper presents an approach that places sensors on patients and household objects, overcoming this limitation. An accelerometer and the transmitter component of a radio-frequency proximity sensor are attached to objects; the receiver component is attached to the arm of interest. The receiver triggers an on-board radio-frequency identification tag to signal proximity when that arm is within 23 cm of an instrumented object. In benchmark testing, this system detected perfectly which arm was used to move the target object on 200 trials. In a laboratory study with 35 undergraduates, increasing the amount of time target objects were moved with the arm of interest resulted in a corresponding increase in system output (p <.0001). Moreover, measurement error was low ( 2.5%). The results support this system s reliability and validity in individuals with unimpaired movement; testing is now warranted in stroke patients. iii

4 TABLE OF CONTENTS Page ABSTRACT... iii LIST OF TABLES... vi LIST OF FIGURES... vii CHAPTER 1 INTRODUCTION...1 Specific Aims...2 Aim Aim Aim Aim BACKGROUND...5 Measurement of Arm Use in Daily Life...5 Accelerometry Research...7 Radio Frequency Identification Systems...8 Movement Sensor...10 Proximity Sensor Research...10 RFID System Software SYSTEM DESIGN...13 Clinical Objective...13 Movement Sensor...14 Selection Constraints...14 Accelerometer...14 Aim 1: Design of the Prototype Proximity Sensor...15 Design Constraints...16 Design Methods...16 Aim 2: Design of the RFID System Software...22 Design Constraints...23 Design Methods...23 Aim 3: Benchmark Testing of the Prototype RFID System...29 Performance Standards for Movement Sensor Testing...29 Performance Standards for Proximity Sensor Testing...30 iv

5 Performance Standards for the Prototype RFID System Testing...31 Aim 4: Reliability, Validity, and Accuracy Testing...31 Research Questions...32 Study Design SENSOR-ENABLED RFID SYSTEM FOR MONITORING ARM ACTIVITY IN DAILY LIFE SENSOR-ENABLED RFID SYSTEM FOR MONITORING ARM ACTIVITY: RELIABILITY AND VALIDITY CONCLUSION AND DISCUSSION...73 Design Considerations...73 Study Limitations...75 Implications and Future Studies...77 REFERENCES...78 APPENDIX 1 IRB APPROVAL LETTER ACTIVE RFID TAG SPECIFICATIONS RFID READER SPECIFICATIONS ACTIGRAPH GT1M ACCELEROMETER SPECIFICATIONS RFID TRACKING SOFTWARE VB.NET CODE RFID DATA PROCESSING SOFTWARE VB.NET CODE COMPUTER METRONOME VB.NET CODE v

6 Table LIST OF TABLES Page SENSOR-ENABLED RFID SYSTEM FOR MONITORING ARM ACTIVITY: RELIABILITY AND VALIDITY 1 Time (S) Objects were Manipulated with the Right Arm per SERSMAA Output (N = 34)...66 vi

7 LIST OF FIGURES Figure Page 1 ActiGraph GT1M Accelerometer Sensor Prototype Low-frequency Radio Transmitter Circuit Diagram Prototype Low-frequency Radio Transmitter Breadboard Testing Setup Prototype Board for Proximity Sensor Transmitter Prototype Low-frequency Radio Receiver Circuit Diagram Prototype Low-frequency Radio Receiver Breadboard Testing Setup Prototype Boards for Proximity Sensor Receiver Prototype Boards in Enclosures for Proximity Sensor Prototype Low-frequency Radio Proximity Sensor RFID Patient Tracking System Hardware Setup RFID Tracking Software Flowchart RFID Tracking Software User Interface Sample Proximity Status File (Left) and Movement Sensor File (Right) RFID Data Processing Software Flowchart RFID Data Processing Software User Interface Combined Data File of Movement and Proximity Sensor Data Summary Variables Data File Object Setup for Test 1 and 2 (Left) and Hairbrush Manipulation (Right) Computer Metronome Program User Interface...33 vii

8 SENSOR-ENABLED RFID SYSTEM FOR MONITORING ARM ACTIVITY IN DAILY LIFE 1 Sketch of SERSMAA Prototype Low-frequency Radio Transmitter Circuit Block Diagram Low-frequency Radio Receiver Circuit Block Diagram Sensitivity of Proximity Sensor Sensitivity of Movement Sensor Performance of SERSMAA System...46 SENSOR-ENABLED RFID SYSTEM FOR MONITORING ARM ACTIVITY: RELIABILITY AND VALIDITY 1 Sketch of SERSMAA Prototype Low-frequency Radio Transmitter Circuit Block Diagram Low-frequency Radio Receiver Circuit Block Diagram Sensitivity of Proximity Sensor Sensitivity of Movement Sensor Setup of Objects for Test 1 and viii

9 CHAPTER 1 INTRODUCTION The ultimate goal of upper-extremity physical rehabilitation after stroke is to improve how well and how much the stroke survivor uses the more-impaired arm in the community [1]. This important outcome, however, has been assessed largely by self-report, which means therapists have to rely on patients judgments about use of their moreimpaired arm. This raises concerns because patients self-evaluation of their arm use may be affected by several types of bias [2]-[7]. To overcome these limitations, Uswatte et al developed an accelerometry system for objectively measuring amount of more-impaired arm movement in daily life [8]. Since this system measures all arm movement over the recording period, an important shortcoming is that it cannot discriminate between functional and non-functional use of the impaired arm. Radio frequency identification (RFID) systems hold the promise of providing a relatively unobtrusive, inexpensive, and practical means of obtaining data to discriminate between functional and non-functional upperextremity movement when the subject is at home and outside the observational scope of experiments. Currently, RFID systems are used commercially to track inventory in retail outlets and for the manufacturing industry, track packages by shipping companies, and track the location of patients and medical equipment in hospitals [9]-[14]. In the present state of the RFID system tracking technology, it cannot capture movement of the tagged object and also it cannot identify whether the object was moved by the person or more specifically by which arm of that person. This study was designed to develop a new objective measure of upper-extremity ac- 1

10 tivity and address the issue of discriminating between functional and non-functional upper-extremity movement. This was achieved by pairing RFID system with sensors to identify and measure movement of household objects, monitor the patient s moreimpaired arm, and finally to associate this arm to the manipulated objects. Accelerometers were used to monitor the movement of the household objects. A prototype proximity sensor was designed to track and monitor the activity of the more-impaired arm with respect to the household objects. When the more-impaired arm was used to manipulate the object, the proximity sensor was triggered indicating the use of that arm, and the movement sensor on that object was also triggered simultaneously indicating the manipulation of that object with that arm. The new RFID system was able to provide information on which household objects were manipulated, whether these objects were manipulated by the patient s more-impaired arm, and finally, when and how much these household objects were manipulated. The goal of designing this new RFID system was accomplished by the following specific aims of the study. Specific Aims Aim 1 To design and build a prototype low-frequency radio transmitter and receiver pair to work as a proximity sensor. A. Design prototype circuits for radio transmitter and receiver B. Evaluate breadboard designs for optimal performance C. Build circuit boards with enclosures for the final prototype sensor 2

11 Aim 2 To design and develop the RFID system software for real-time sensor tracking and offline data processing. A. Program interface to permit user to set-up system for recording data B. Program modules for data acquisition from RFID readers and active RFID tags C. Program modules for providing real-time tracking information on the movement and handling of tagged objects D. Program modules for saving data and for offline processing and analysis Aim 3 To conduct benchmark testing of the prototype RFID system. A. Evaluate whether the movement sensor meets necessary performance standards B. Evaluate whether the prototype low-frequency radio proximity sensor meets necessary performance standards Aim 4 To test how reliably, validly and accurately the prototype RFID system captures movement of household objects by healthy adults and amount of time the designated arm is used. The end result of this study was a prototype RFID system which was able to track whether the subject used the more-impaired arm or arm of interest to manipulate the household objects (real-time proximity sensor, design and development (Aim 1) de- 3

12 scribed in Chapter 3), which object was manipulated (ID of movement sensor for that object), when they were manipulated (time stamp on each trigger event), and for how long the objects were manipulated (synchronous time for which the proximity sensor and movement sensor stay ON, prototype system software developed to collect and analyze the data from the movement and proximity sensors (Aim 2) and is described in Chapter 3). Benchmark testing of the system (Aim 3) was carried out (Chapter 4) to test the design performance standards set by the clinicians for measuring and monitoring arm use. Reliability, accuracy and validity of the system were evaluated by testing the system with normal Psychology 101 college students (Aim 4) in three test procedures (Chapter 5). 4

13 CHAPTER 2 BACKGROUND Measurement of Arm Use in Daily Life Measuring arm use in daily life is important because evidence and theory indicate that laboratory measures of impairment and function do not predict how neurorehabilitation patients actually function in daily life. The learned nonuse formulation provides a wellsupported explanation of how use of part of a body in daily life and motor capacity or ability as measured by laboratory tests for motor performance can diverge. E. Taub showed in monkeys with a single forelimb from which sensation is surgically removed that such animals do not use this deafferented limb even though they have adequate motor innervation to do so [15]. This excess motor deficit develops soon after the deafferentation surgery because the animal cannot adequately coordinate movement of the deafferented limb due to depressed activity of the still intact motor neurons innervating it. The deficit in coordination, which is temporary, leads to failures in use of the deafferented limb (e.g., dropping food that the animal attempts to pick up) that punish the animal for attempts to use it. Punishment results in suppression of further attempts to use the deafferented limb. At the same time, the animal can operate on its environment successfully using its other limbs so this compensatory pattern of behavior is strengthened. These two processes result in continued nonuse of the deafferented limb even after the still intact motor neurons innervating it return to normal activity levels several weeks after surgery. A parallel set of conditions prevails after stroke. Stroke survivors typically endure a severe loss of function on one side of their body that gradually resolves over time. 5

14 Over the course of their recovery, stroke survivors are also able to use the other side of their body effectively to accomplish important tasks. Not surprisingly then, learned nonuse is also observed in stroke survivors and survivors of other types of brain injury [1], [15]-[17]. The implication of the learned nonuse formulation for measuring motor function in stroke survivors and other rehabilitation populations is that it is necessary to assess both the motor capacity of a body part and its actual use in daily life. Failure to measure actual use of a body part has several adverse consequences. First, rehabilitation interventions might be deemed efficacious when they effect a change in motor ability as assessed in the laboratory on a performance test but do not improve motor function in the domain that matters most, i.e., patients daily lives. Second, a bias over time that might skew research efforts so that little attention is given by investigators to modify actual real-world use of an impaired-extremity. Third, the deficits in real-world use might be under-diagnosed and under-treated [18]. Over the last two decades, therefore, substantial effort has been spent on developing measures of function in daily life (e.g., most recently the Late Life Function and Disability Instrument [19]). In the area of upper-extremity neurorehabilitation, Taub and coworkers developed a structured interview for the purpose of measuring real-world moreimpaired arm use in adults after stroke, known as the Motor Activity Log (MAL) [16], [20], [21]. Patients rate their more-impaired arm use on quality (how well scale) and amount (how much scale) for 30 upper-extremity activities of daily living over a specified period (e.g., the past week). These two scales rate arm use on a scale of zero to five, where zero is no-use and five is normal use. 6

15 The MAL measure for arm use is a self-report measure. As with all self-report measures, MAL is also subject to biases inherent in subjective reports [2]-[7]. Recall bias is one of the most common biases occurring in self-reports, where patients self-rating is biased or faulty due to inaccurately remembering the performed tasks. To address the issue of biases in self-report measures, several researchers have used accelerometry as an objective measure for the amount of more-impaired arm activity in the real life home environment of the patient. Accelerometry Research Several research groups, including the CI Therapy Research Group at University of Alabama at Birmingham (UAB), have employed accelerometers to measure the amount of physical activity or movement of upper-extremity in older adults in the community to combat the concerns related to self-report measures [8]. In the work by the UAB CI Therapy Research Group, stroke survivors with mild-to-moderate motor impairment of the arm on the more-affected side of the body were asked to wear accelerometers on their arms during all waking hours for 2 days before and after upper-extremity physical rehabilitation or a corresponding no-treatment period. Bi-axial accelerometers inserted in arm bands were worn by the subjects, which measured their arm movements for the activities performed. Data from the accelerometers were extracted and analyzed by computer programs. The results showed that the accelerometer measure of upper-extremity movement was strongly correlated with the amount of impaired-arm use in daily life after stroke (r, range =.5-.94) and is a reliable and valid measure of upper-extremity rehabilitation outcome [8]. Since accelerometers measures all arm movements, however, they can not dis- 7

16 criminate whether a given movement is functional or non-functional and can not identify what tasks were performed. A functional movement of the arm is one which can be associated with a task involving a household object, whereas a non-functional movement is a random movement or non-task related movement like when the arm is moved while walking. Thus, there is a need of a system which will not only measure the movement of the arm, but measure this movement with respect to real life tasks. Radio Frequency Identification Systems RFID systems have the potential of identifying and measuring functional movements of the tagged arm, by capturing movements of tagged household objects with respect to that arm. To date, RFID technology has been used to track hospital equipment and patients. RFID system possesses the potential for tagging objects in a building and monitoring its position and location from a central area. This way hospital staff can keep track of their equipment, and in case of an emergency, equipment can be tracked and located quickly, reducing delays to providing service. RFID has also been shown to be effective in keeping track of patients in the hospital if they wander away. Patients are asked to wear tagged wrist bands which communicate their location to the monitoring staff at all times. This way the RFID system helps hospitals to monitor a patient s location and notify hospital staff or caregivers of the patient s location or any particular hazard situation. RFID systems have demonstrated the potential of location tracking of objects as well as humans in the same area. When attached to humans, RFID tags help administrators and security personnel keep track of or locate personnel in a particular building or area when the need arises [9]-[14]. Radio-frequency identification is achieved by attaching a RF tag to a product, animal, 8

17 or person for tracking and identification using radio waves. RF tags can be read from several meters away and does not need to be in the line of sight of the RF reader. Tags have unique identifier microchip similar to barcode and magnetic strip. RF tags when attached to inventory helps uniquely identify the status and location of that object. Most RFID tags consist of at least two parts; a) an integrated circuit for storing and processing information, and also modulating and demodulating a RF signal, and b) an antenna for receiving and transmitting the RF signal. RFID tags are usually found in three basic types; a) active RFID tags, which contain a battery and can transmit signals autonomously, b) passive RFID tags, which have no battery and require an external power source to provoke signal transmission, and c) battery assisted passive (BAP), which requires an external power source to wake-up but has good read range using forward link capability. The RFID tag data is sent to the RF reader which in turn interfaces to the monitoring computer. The embedded signal conditioning circuitry in the RF reader deciphers this RF tag data and sends real-time information to the monitoring computer. The user can use this information in several applications raging from controlling access to restricted areas, tracking assets through a facility, or updating inventory counts in a warehouse. After an arm rehabilitation therapy, clinicians need to know how much the moreaffected arm was used in the home environment and whether the use has increased from pre- to post-therapy. Also, they need to know the use of this more-affected arm with the manipulation or movement of the objects in the home environment which directly associates the use of that arm with tasks involving those objects, and thus evaluate the functional use of the more-affected arm. At the current state-of-the-art, an RFID system 9

18 can only identify objects and cannot capture movement of these objects in the home environment. Also, the RFID system lacks the capability of detecting the proximity of the person s arm that was used for a specific task. This gives rise to the need for pairing an RFID system with a movement sensor for identifying the movement of the particular object. A proximity sensor is also needed with this RFID system to identify which arm was used to manipulate the household objects. Movement Sensor A movement sensor can measure the amount of movement when attached to the moving object. Accelerometers have been successfully used in the past to measure amount of movement with respect to physical activity in the home environment of the patients [8]. Accelerometers when paired with RFID system can provide a richer picture of how much the household objects were manipulated and also ID the objects. In addition to the movement sensor the RFID system needs to identify which arm is moving the object. Proximity Sensor Research To measure use of a particular arm with respect to manipulation of objects, a proximity sensor can be used to sense the proximity of subject s arm to household objects. The proximity sensor can sense whether the subject manipulates the object with the arm of interest and will allow the RFID system assessment of manipulations for that particular arm. The two most common technologies in proximity detection include infra-red proximity sensors and magnetic proximity sensors. Infra-red proximity sensors have an infra- 10

19 red emitter and a receiver on the other end. In such proximity sensors, the infra-red emitter needs to be in direct line of sight of the receiver to effectively operate. In most cases the emitter and the receiver are mounted together on the same board and pointing to the same direction. In this case the infrared proximity sensor acts as a distance sensor or an obstruction sensor. Again the direction is restrictive and operates only on the direction it is facing [23], [24]. The infra-red emitter signal is also easily obstructed by any opaque object like a book or a coffee mug, and will restrict identifying accurately whether the particular arm was used to manipulate the object. Magnetic proximity sensor carried by ActiveWave Inc [22] senses proximity of a magnetic field. To have a particular arm detected using a magnetic proximity sensor the subject would need to wear a magnet on that particular arm with the proximity sensor attached to the object. Since the magnetic field of a magnet drops exponentially, the use of a very strong magnet makes the size and weight of the magnet unrealistic to be used on the wrist or finger of a person or even attach the magnet on an object. Orientation of the magnet with respect to the sensor affects proximity detection, and the sensor fails to detect proximity if the magnet is not facing the sensor. The above issues with magnetic proximity sensors restrict its use to identify proximity of a particular arm. The use of a wireless radio frequency based proximity sensor appears to have the ability to meet the design and performance standards. Currently, however, there are no off-the-shelf wireless proximity sensors available to be adapted for this study. Therefore, the design and development of a prototype wireless proximity sensor can help in the detection of the proximity of the particular arm to the manipulated objects. This sensor when paired with the RFID system and movement sensor can provide information about 11

20 the use of the arm with respect to the manipulation of the objects in the home environment. RFID System Software Off-the-shelf RFID software development kit (SDK) provides users to design and integrate RFID tags and other sensors for individualized applications. The SDK provides several platforms such as C# (C-sharp) or VB.NET (Visual Basic.NET) for users to write their custom application program codes for integration of RFID tags and sensors deployed in the observational or tracking area. Industry applications for software include tracking inventory and synching data with inventory database for keeping log of products. Software developed for hospital applications track RFID tags attached to patients or hospital equipment and display tracking information to clinicians or send data to mobile devices if needed. Since each application is unique and based on the tracking requirements of that project or application, off-the-shelf software cannot fulfill the tracking requirements of this project. Therefore, the development of custom RFID system software interfacing with such a SDK package can help in tracking the sensors paired with the RFID system. Also, additional software is required with data processing capabilities and not interfaced to the SDK to process the sensor data files offline for computing the summary variables required for evaluating the tracking sessions. 12

21 CHAPTER 3 SYSTEM DESIGN Clinical Objective The first application of the prototype RFID system will be to evaluate and measure the changes in arm use before and after upper-extremity rehabilitation therapy. This will be achieved by tracking the movement of a sample of household objects 2-3 days pre- to post-therapy and associating this movement to the tagged arm. Arm use will be evaluated based on the manipulation of this tagged sample of household objects. The objective will not be to tag all objects in the household. Since the primary goal is to capture pre- to post-rehabilitation changes in everyday arm use, detecting changes in commonly used household objects will answer this question adequately. Another consideration is that one rarely attempts to measure all instances of a target behavior in clinical research. The typical strategy is to collect data on a representative sample of the target behavior. The prototype RFID system will primarily address the issue of discriminating between functional task related arm movements and non-functional arm movements; non-task related functional arm movements will not be captured by this system. Task-related functional arm movements are of primary interest because one of the major goals of rehabilitation is to improve how well and how often patients accomplish tasks in their everyday life. The design of the prototype RFID system was achieved by pairing the RFID system with movement sensors to capture the movement of household objects and proximity sensor to detect the arm which was used to manipulate these objects. This system when implemented in the home environment pre- and post-rehabilitation therapy of a stroke survivor 13

22 can provide much richer information on the use of the more-impaired arm in daily life and enable the clinicians to make a better assessment of the effectiveness of upperextremity rehabilitation therapy. Movement Sensor A movement sensor was paired with the RFID system to monitor and measure the movement of the household objects when manipulated by the tagged arm. Selection Constraints 1. The movement sensor was required to measure at least 8 cm movement. 2. The sensor should measure this movement continuously without having a delay in reset or readjustment. 3. The movement sensor selected should measure movement in all axes of movement accurately. 4. The movement sensor was required to measure continuously for at least 12 hours. Accelerometer Off-the-shelf activity monitor ActiGraph ( GT1M bi-axial accelerometer was selected based on the above selection constraints. The GT1M accelerometer movement sensor was used to tag household objects and paired with the RFID system to sense movement of these objects when manipulated by the subject. The movement sensor is sensitive to change acceleration of the attached object ranging in magnitude from approximately 0.05 to 2 G and thus senses movement of the object. The accel- 14

23 eration measurements occur for a user specified interval of time called an epoch and stored on-board 1 MB flash memory. The GT1M is compact with a weight of 27 gm; dimension of 3.8 cm x 3.7 cm x 1.8 cm, and each sensor has unique sensor ID. The selection constraints were met by rigorous laboratory testing of the GT1M accelerometer (Aim 3 and Chapter 4). Figure 1 shows a picture of the GT1M sensor and manufacturer s technical specification information is attached in Appendix cm x 3.7 cm x 1.8 cm Weight = 27 gm Figure 1 ActiGraph GT1M Accelerometer Sensor Aim 1: Design of the Prototype Proximity Sensor The design and development of a prototype proximity sensor and pairing it with RFID system addresses the need of detecting which arm was used in manipulating the objects in the home environment. The proximity sensor senses the proximity of the tagged arm with respect to the tagged object. When used in conjunction to the movement sensor, it will enable the prototype RFID system to not only detect the movement of the objects but also associate this movement to the arm manipulating these objects. 15

24 Design Constraints 1. The prototype proximity sensor was required to measure the proximity of the more-impaired arm within 23 cm from the object. 2. The proximity sensor should perform with at least > 90% accuracy and precision regardless of the angle of approach (x, y or z axis). 3. The proximity sensor should perform with at least > 90% accuracy and precision regardless of the type of object manipulated. 4. The proximity sensor was required to send data signals from a distance of about 35 m, and regardless of household obstruction from furniture, walls or room dividers. Design Methods Wireless technology was used for the design of the prototype proximity sensor as radio signals do not have orientation limitation as seen in present proximity sensor technologies. The prototype wireless proximity sensor was designed to work in a pair with two components; a transmitter component to be attached to the objects, and a receiver component to be worn on the more-impaired arm of the patient. A low-frequency 10.7 KHz signal was selected for the operation of the sensor to maintain or restrict the sensor to a small detection range of about < 23 cm. The receiver was design to have an embedded active RFID tag for relaying the proximity status to the monitoring computer. Active RFID tags can send signals at about 45 m from the RF reader and monitoring computer [22]. The above design selections ensure to meet the design constraints set for the proximity sensor. Prototype low-frequency radio transmitter. Figure 2 shows the circuit diagram of the 16

25 prototype proximity sensor transmitter. The transmitter has two functional parts; a square wave oscillator and the low-frequency radio transmitter. The oscillator circuit was designed to provide a constant trigger of 30 Hz square wave signal to be transmitted using the low-frequency radio transmitter. The 30 Hz square wave signal was designed using an integrated circuit NE555 timer chip. The frequency of the square wave oscillations was configured by the following formula; F Oscillator = C 1.( R1 + 2R2)...1 Where, C1, R1 and R2 are the circuit components shown in Figure 2, the circuit diagram for the low-frequency radio transmitter. After several iterations, the radio transmitter was set to send the 30 Hz radio signals at 10.7 KHz. This low-frequency was selected to maintain a low detection range for the proximity sensor of < 23 cm. The transmitter was designed using an n-p-n transistor 2N3904. The frequency of the transmitter was calculated using the following formula; F Transmitter = 2π. 1 L1. C4...2 Where, L1 and C4 are the inductor and capacitor used to configure the frequency of the low-frequency transmitter part of the circuit in Figure 2. Figure 3 shows the testing of the low-frequency transmitter circuit on a breadboard. After successful breadboard testing, the circuit components were mounted on prototyping board shown in Figure 4. 17

26 Figure 2 Prototype Low-frequency Radio Transmitter Circuit Diagram Oscillator Circuit Low-frequency Transmitter Circuit Figure 3 Prototype Low-frequency Radio Transmitter Breadboard Testing Setup Figure 4 Prototype Board for Proximity Sensor Transmitter Prototype low-frequency radio receiver. Figure 5 shows the circuit diagram for the low-frequency receiver component of the proximity sensor. The receiver has two func- 18

27 tional parts; the low-frequency radio receiver circuit, and the frequency switch. The lowfrequency radio receiver was designed using two n-p-n transistor 2N3904, and tuned to the same 10.7 KHz frequency as the transmitter. The 10.7 KHz frequency was set using the inductor L1 and capacitor C3 in the receiver circuit in Figure 4. The receiver was tuned to receive the 30 Hz square wave signal transmitted by the proximity transmitter to sense the proximity of the transmitter at < 23 cm. The frequency switch was designed using an integrated circuit LM2907. The frequency switch in the receiver circuit turns ON and OFF based on the presence and absence of the 30 Hz square wave oscillating signal. The triggering frequency of the switch was configured by the following formula; F Switch = 1 2π. R11. C7...3 Where, R11 and C7 are the resistor and capacitor in the switch part of the circuit in Figure 5. The frequency switch triggers an ActiveWave RFID compact tag [22] which in turn sends proximity status signals to the RF reader and the monitoring computer. Appendix 3 gives the specification of the ActiveWave Inc. active RFID tag and Appendix 4 give the specification for the RF reader. Figure 6 shows the breadboard circuit testing setup for the prototype receiver. After successful breadboard testing of the proximity receiver circuit, the circuit components were mounted on a prototyping board shown in Figure 7. 19

28 Figure 5 Prototype Low-frequency Radio Receiver Circuit Diagram 20

29 Active RFID Tag Low-frequency Receiver Circuit Frequency Switch Figure 6 Prototype Low-frequency Radio Receiver Breadboard Testing Setup Figure 7 Prototype Boards for Proximity Sensor Receiver Figure 8 shows the prototype boards of the proximity transmitter and receiver components in enclosures for the final prototype sensor. Figure 9 gives the dimensions of the prototype proximity sensor. Figure 8 Prototype Boards in Enclosures for Proximity Sensor 21

30 7.8 cm x 3.8 cm x 2 cm Weight = 50 g 7.4 cm x 6.1 cm x 2.4 cm Weight = 95 g Figure 9 Prototype Low-frequency Radio Proximity Sensor Operation of the proximity sensor. In proximity of the radio transmitter (on object) of < 23 cm, the radio receiver (on the arm) senses the proximity of the transmitter and triggers the embedded active RF tag to alert the RFID system that the particular arm is in proximity to the object. Aim 2: Design of RFID System Software ActiveWave Inc. supplied the RF reader and active RFID tag for the integration of the proximity sensor. The SDK package for this system allows the user to write custom software applications in VB.NET for individualized applications. In this project, custom RFID system software was developed for tracking the proximity sensor which was used to track the proximity of the arm to the object manipulated. The movement sensor could not be tracked by this software and the data needed to be downloaded from the device onboard memory offline for further integration and processing. Additional VB.NET software was developed with data processing capabilities to process the sensor data files off- 22

31 line for computing the summary variables required for evaluating the tracking sessions. Design Constraints 1. The RFID tracking software would setup a communication channel between the computer and the RF reader using LAN networking. 2. The RFID tracking software was required to track real-time proximity status and send alert messages when the more-impaired arm was in proximity to a particular object. 3. The RFID tracking software would save tracking data in text files for offline processing and data analysis. 4. The RFID data processing software was required to perform offline data processing of the movement and proximity sensor data text files and produce reports from the tracking data. Design Methods To meet the design constraints, the RFID system software was designed in two parts; a) communicate with the RF reader and capture RF tag information, and save data in text files (RFID tracking program), and b) process the data text files offline and compute summary variables (RFID data processing software). RFID system hardware setup. A local area network (LAN) connection was setup between the RF reader and the computer via a network switch. The switch provides an interface for the RF reader and the computer to communicate with each other in the LAN and ensure reliable data transmission. Figure 10 shows the hardware setup of the system. 23

32 Figure 10 RFID Patient Tracking System Hardware Setup RFID tracking software. Figure 11 gives the software algorithm flowchart which illustrates the flow of control within the RFID tracking software. The RFID tracking software was designed in VB.NET with the following functional modules (M1-5). M1. User input for accepting data file name. M2. Setup and initiate communication with the RF reader using Ethernet protocol for collecting data. M3. Detect proximity of low-frequency transmitter (particular arm which is being used to move the object) based on low-frequency receiver and RFID active tag signals. M4. Display real-time tracking information in the form of on-screen messages. M5. Store tracking data in text files for offline processing and data analysis. Figure 12 shows the software interface screen for the developed VB.NET program (software code is attached in Appendix 5). Figure 13 shows a sample proximity data file saved by the RFID tracking software and movement sensor file downloaded to a text file from the accelerometer movement sensor. Figure 13 also shows highlighted regions where movement sensor and proximity sensor triggered simultaneously with overlapping time stamp ID. 24

33 Start Initialization M1: User Input Specify data file name M2 Initiate RF Reader detection and IP network connection M2: RF Reader detection & network setup module M3: Is Tag Triggered? YES M4: Audio Signal + On-screen Message NO Data Manipulation M3: Is STOP initiated? M5: Record data in Text Stop YES Figure 11 RFID Tracking Software Flowchart 25

34 Figure 12 RFID Tracking Software User Interface RFID data processing software. RFID data processing software was developed in VB.NET for offline processing of the movement and proximity sensor data text files. This software was designed to compute summary variables from the data files. Summary variables like total time in movement by the more-impaired arm for manipulating the objects, and also time in movement in manipulating each of the household objects. Figure 4 describes the flow chart schematic in the data processing steps. The software was designed with four modules; select data files for processing, transform accelerometer data files, combine all the data files in one file matching the time stamp, and compute summary variables. Accelerometer data files were transformed from two axis data to single axis time stamped data file and also threshold filtered to transform the data to binary values (values > 2 = 1 and values 2 = 0). The threshold value of 2 26

35 was selected as a lowest acceleration value above which data was coded as presence of movement and no movement for values 2 or below. Figure 15 shows the user interface screen for the VB.NET RFID data processing software (software code in Appendix 6). Figure 13 Sample Proximity Status File (Left) and Movement Sensor File (Right) RFID data processing software combines the transformed accelerometer and proximity status files in a combined file by matching the records in both files by the timestamp as an ID for each record. Figure 16 shows the combined data file with the highlighted region for the arm activity for object 1 and object 2. The summary variables module computes the time in movement for each object by scanning the combined data file for synchronous triggering of both the movement and proximity sensor. Figure 17 shows the summary variables file created by the RFID data processing software. The summary data was analyzed by SAS 9.1 statistical software. 27

36 Start Select data files for processing (Figure 13) Module to transform accelerometer data files Module to combine accelerometer and proximity status files using time stamp ID (Figure 16) Module to compute time in movement of the objects and total time in movement by the tagged arm (Figure 17) Stop Figure 14 RFID Data Processing Software Flowchart Figure 15 RFID Data Processing Software User Interface 28

37 Figure 16 Combined Data File of Movement and Proximity Sensor Data Figure 17 Summary Variables Data File Aim 3: Benchmark Testing of the Prototype RFID System This section is aimed at testing the sensitivity and specificity of the prototype RFID system. The performance standards for the sensors define the requirements of the sensor to perform satisfactorily based on the design constraints set earlier. Testing ensures that the sensors and the system meet the characteristics and specifications required for reliably measuring the amount of movement of the object and also correctly identify the arm. 29

38 Performance Standards for Movement Sensor Testing Benchmark testing was performed (Chapter 4) to evaluate the following performance standards and behavior necessary for the accelerometer movement sensor to meet the goals and specific aims of the study. 1. If the tagged object is moved at least 8 cm, the movement sensor will detect a change in acceleration with an accuracy of 98%. 2. The movement sensor detects movement with at least 98% sensitivity for x, y and z axis of movement. 3. The movement sensor detects movement with at least 98% sensitivity for 3 s inter-movement interval. 4. If the tagged object is at rest, the movement sensor does not trigger or register any movement signals. Performance Standards for Proximity Sensor Testing Benchmark testing was performed (Chapter 4) to evaluate the following performance standards and behavior necessary for the low-frequency radio proximity sensor and active tag to meet the goals and specific aims of the study. 1. The proximity sensor (receiver) senses proximity of the tagged object (transmitter) within 23 cm and with at least 98% accuracy. 2. The receiver-tag triggers 98% of the time regardless of the approach angle to the object transmitter, or the type of object the transmitter is attached to. 3. The proximity sensor detects 98% of the time for 1 s inter-movement interval. 4. The proximity sensor does not trigger when the receiver and transmitter compo- 30

39 nents are placed > 23 cm away from each other. 5. The RF signal is not obstructed by room dividers, doors and furniture within the ~35 m radius signal transmission range. Performance Standards for the Prototype RFID System Testing Benchmark testing was performed (Chapter 4) to evaluate the following performance standards and behavior necessary for the prototype RFID system to meet the goals and specific aims of the study. 1. The RFID system detects at least 98% of the time the tagged arm is used to either hold the object or even in proximity of < 23 cm of the object. 2. The RFID system detects at least 98% of the time when the tagged arm is not used to manipulate the object or other arm is used or another person manipulates the object. 3. The RFID system computes time in movement of the object and the arm used to manipulate this object with at least 98% accuracy. 4. The RFID system does not detect the tagged arm when that arm is > 23 cm from tagged objects. Aim 4: Reliability, Validity, and Accuracy Testing This study was aimed at testing the prototype RFID system for validity of measuring varying amounts of object movement, whether the system can measure these movement times accurately, whether the system can reliably measure these movements, and also test the system whether it can correctly identify which arm was use to manipulate the objects. 31

40 Research Questions How reliably, validly and accurately does the prototype RFID system measure movement of household objects by healthy individuals in the laboratory? A. To test whether the RFID system can reliably, validly and accurately detect how many times and which object has been manipulated? B. To test whether the RFID system can reliably, validly and accurately measure how long an object has been manipulated? C. To test whether the RFID system can reliably, validly and accurately determine that the objects were moved by the designated arm or the other one? Study Design College students were recruited in a study to evaluate reliability, validity and accuracy of the prototype RFID system paired with movement and proximity sensors. Each student was instructed to move five household objects (cup, book, remote, hairbrush, telephone receiver) in three test procedures. The three test procedures were designed to test the prototype RFID system (Chapter 5) for varying movement times (Low, Medium, High) and also varying number of objects moved (1, 3, 5). Proximity receiver was attached to the student s right arm and the objects were tagged with proximity transmitter and accelerometer movement sensor. Figure 18 shows the setup of the 5 objects for Test 1 and 2. The objects were placed 28 cm apart and the objects were moved between the starting and target location 14 cm apart. The objects were moved in rhythm to a 40 BPM computer metronome. Figure 19 shows the user interface for the VB.NET computer metronome program (software code attached in Appendix 7). 32

41 14 cm 28 cm 14 cm Figure 18 Object Setup for Test 1 and 2 (Left) and Hairbrush Manipulation (Right) Figure 19 Computer Metronome Program User Interface 33

42 CHAPTER 4 SENSOR-ENABLED RFID SYSTEM FOR MONITORING ARM ACTIVITY IN DAILY LIFE by JOYDIP BARMAN, MEMBER, IEEE, GITENDRA USWATTE, NILANJAN SARKAR, TOURAJ GHAFFARI, AND BRAD SOKAL Conf Proc IEEE Eng Med Biol Soc, Boston, MA, 2011, pp Copyright 2011 by Joydip Barman Used by permission Format adapted for dissertation 34

43 Abstract After stroke, capacity to carry out tasks in the treatment setting with the moreaffected arm is a poor index of actual use of that extremity in daily life. However, objective methods currently available for monitoring real-world upper-extremity use only provide information on amount of activity. These methods, which rely on movement sensors worn by patients, do not provide information about type of activity (e.g., functional vs. nonfunctional movement). The benchmark testing reported here evaluated an approach that involves placing sensors on patients and objects. An accelerometer and the transmitter component of a prototype radio frequency proximity sensor were attached to household objects. The receiver component was placed on the experimenter s right arm. This device triggered an on-board radio frequency identification tag to signal proximity when that arm was within 23 cm of the objects. The system detected > 99% of 6 cm or greater movements of objects. When handling of objects by the right or left arm was determined randomly, 100% of right arm trials were detected. No signals were recorded when objects were at rest or moved by the left arm. Testing of this approach, which monitors manipulation of objects (i.e., functional movement), is now warranted in stroke patients. Keywords Manuscript received March 26, This work was supported by the Department of Psychology at the University of Alabama at Birmingham (UAB), and NIH grant HD J. Barman is with the Departments of Biomedical Engineering and Psychology, UAB, Birmingham, AL USA (phone: ; joydip@uab.edu) G. Uswatte is with the Departments of Psychology and Physical Therapy, UAB, Birmingham, AL USA ( guswatte@uab.edu) N. Sarkar is with Department of Mechanical Engineering, Vanderbilt University, Nashville, TN USA ( nilanjan.sarkar@vanderbilt.edu) T. Ghaffari is with ActiveWave Inc., Boca Raton, FL USA ( touraj@activewaveinc.com) B. Sokal is with the Department of Psychology, UAB, Birmingham, AL USA ( bhsokal@uab.edu) 35

44 Radio frequency identification, RFID, sensors, acceleration, proximity, arm, activity, ambulatory monitoring, rehabilitation, stroke I. INTRODUCTION More than 650,000 survive strokes annually in the United States [1]. Persistent impairment of the arm on the more-affected side of the body afflicts between 55% and 75% of survivors [2] and is associated with diminished health-related quality of life [3]. Advances in methods to assess and treat more-affected arm impairment after stroke, therefore, have the potential to improve the lives of many. Well-known models of disability and data indicate that laboratory measures of function poorly index how stroke survivors actually use their more-affected arm in daily life [4]. Therefore, substantial effort has been spent on developing real-world measures of arm function. Most of these tests, however, rely on self-report [4]. Researchers have objectively measured amount of arm activity in the community by placing accelerometers on stroke survivors [5]. These techniques, however, cannot discriminate whether a given arm movement is functional or non-functional and cannot identify what tasks were performed. More complex activity monitors, such as Inertial Measurement Units, hold promise for making such discriminations but to date have been shown only to index quality of arm movement after stroke on a standardized motor test in the laboratory [6], [7]. This paper describes design and benchmark testing of a system of radio frequency identification (RFID) tags paired with proximity and movement sensors for measuring arm activity. In this approach, movement sensors (i.e., accelerometers) are placed on objects, along with one component of a RF proximity sensor. The other component of the 36

45 proximity sensor is connected to an active RFID tag and placed on the arm of interest. Manipulation of instrumented objects with that arm produces synchronous signals from the movement and proximity sensors, permitting tracking of which objects are handled, when handling takes place, and whether handling is by the person and arm of interest. The proposed approach, thus, can collect much richer objective data than possible now. II. BACKGROUND A. Monitoring of Arm Activity with Accelerometers To overcome the limitations of self-reports, several researchers have employed accelerometers to objectively measure amount of arm activity in stroke survivors in the community [5]. For example, Uswatte et al. [8] asked stroke survivors with mild-to-moderate impairment of their more-affected arm to wear an accelerometer above each wrist during all waking hours for 2 days before and after upper-extremity physical rehabilitation or a corresponding no-treatment period. They found that the ratio of more-affected to lessaffected arm accelerometer recordings was strongly correlated with amount of moreaffected arm use in daily life (r =.74, p <.001). However, since any arm movement produces an acceleration reading, such approaches cannot discriminate whether a given movement is functional or non-functional nor identify what tasks are performed. B. Radio Frequency Identification Systems RFID systems consist of small tags that transmit a unique ID using RF and a RF reader that monitors the status of these tags [9]. Software on a PC connected to the reader processes the RFID signals. Passive RFID tags transmit their ID when they encounter 37

46 the reader s radio waves, whereas active RFID tags, which are battery powered, transmit their ID independently from as far as 85 m [10]. Typical applications involve tracking whether tagged objects are within the range of the reader or not. Examples are monitoring when hospital equipment or a patient leaves a room and monitoring how much merchandise remains in a warehouse [11]. RFID systems have not been used to remotely monitor upper-extremity activity in stroke survivors or other populations, for that matter. III. METHODOLOGY A. Apparatus 1) Sensor-enabled RFID system for monitoring arm activity (SERSMAA). Fig. 1 shows the hardware setup, and how the movement and proximity sensors operate together when an object is manipulated with the arm of interest. Fig. 1. Sketch of SERSMAA Prototype A local area network (LAN) is setup between the PC and RF reader using an Ethernet 10/100 Mbps switch. The switch enables the RF reader and PC to communicate reliably 38

47 over the LAN. A movement sensor (5 cm x 4 cm x 1.7 cm; 37 g) and proximity sensor transmitter (7.8 cm x 3.8 cm x 2 cm; 50 g) are placed on each object. The receiver component of the proximity sensor is connected to an on-board active RFID tag; this assembly (7.4 cm x 6.1 cm x 2.4 cm; 95 g) is attached to the arm of interest. Each movement sensor [12] and the RFID tag [10] possess a unique ID. When the arm of interest approaches an instrumented object, the proximity sensor receiver detects the transmitter s signals, triggering the RFID tag to broadcast an ON signal along with its ID. When the arm withdraws, the proximity sensor receiver no longer detects the transmitter s signals, triggering the RFID tag to broadcast an OFF signal along with its ID. The RF reader relays the proximity status signals to the PC, which runs custom software that processes these signals and stores the output in a text file. If the object is manipulated, the movement sensor records the changes in its acceleration, and stores these values in on-board memory for offline downloading into a text file that includes the movement sensor ID. Custom software processes the proximity and movement sensor text files offline. Synchronous positive values from the proximity and movement sensors indicate that an instrumented object is being moved by the arm of interest. Moreover, analysis of the proximity status and acceleration values, along with their ID and time stamps, permits tabulation of which objects are moved, when they are moved, for how long, and by which arm. 2) Proximity sensor. Fig. 2 and Fig. 3 shows block diagrams of the transmitter and receiver components, respectively, of the prototype RF proximity sensor. As noted, transmitters are attached to objects, while the receiver is attached to the arm of interest. 39

48 Oscillator Circuit Low-Frequency Radio Transmitter Fig. 2. Low-Frequency Radio Transmitter Circuit Block Diagram Low-Frequency Radio Receiver Frequency Switch RFID Active Tag Fig. 3. Low-Frequency Radio Receiver Circuit Block Diagram The RF transmitter sends 30 Hz oscillator signals at a fixed low frequency of ~10.7 KHz. A low-frequency is desirable for sensing proximity of the receiver and transmitter over distances of 1 to 23 cm, i.e., for detecting when the arm of interest is close to an instrumented object. The RF receiver is tuned to the same frequency as the transmitter. The receiver output connects to a frequency switch circuit that turns ON when it reads the 30 Hz signal and turns OFF in its absence. Because the frequency switch output cannot be readily connected to the ActiveWave RFID tag [10], a jerry-rigged solution is used in this prototype. The frequency switch output, instead of firing the tag directly, connects to an electromagnet, which produces a magnetic field when the frequency switch toggles ON. This magnetic field, in turn, activates an ActiveWave magnetic sensor active RFID tag, which sends a signal to the RF reader indicating a change in sensor status. The power sources for both components are 3 V coin cell batteries. 3) Movement sensor. The object movement sensors are ActiGraph GT1M Activity Monitors. GT1M units employ a biaxial accelerometer, which detects 1 g acceleration with a sensitivity of ±10%. Acceleration is sampled at 60 Hz in each axis. These samples are integrated separately for each axis over a user-specified epoch, which in this case was 1 s, and are stored in 1 Mb flash memory [12]. To remove non-functional movement 40

49 (e.g., simply brushing the arm of interest against an object), integral values 1 are set to 0; other values are set to 1 [13]. To generate a single ON/OFF movement signal, the movement status in each epoch is set to OFF only if the threshold-transformed integral values for both axes are 0. Otherwise, movement status is set to ON. Biaxial accelerometers are adequate for monitoring arm activity because manipulation of objects invariably results in movement components in all 3 axes [14]. B. Procedure Benchmark testing was performed under highly controlled conditions in the laboratory to determine whether the sensitivity and specificity of the SERSMAA prototype was adequate, i.e., 98% and >99%, respectively. 1) Proximity sensor testing. The proximity sensor transmitter was attached to the side of a coffee mug using Velcro. The proximity sensor receiver was attached with an elastic band to the right forearm of the experimenter just above the wrist. The mug was placed on a target at the center of several circles of varying radii drawn on a table. Two hundred trials of each test were conducted, except for Test 1a, which had 100. The start and end of trials were marked with beeps emitted by custom software on a PC. a) To determine the range of proximity detection, the experimenter moved his hand along the table top in 1 cm increments every 5 s starting from a target 24 cm away from the mug and ending 20 cm away. Movement was parallel to the y axis of the mug. Proximity sensor status was recorded at each 1 cm increment. b) To evaluate how sensitivity varies with angle of approach, the experimenter placed his hand on a target > 23 cm from the mug. The experimenter then grasped the mug han- 41

50 dle with his right hand, released it, and returned his hand to the target. This movement was conducted parallel to the x, y, and z axes of the mug in separate sets of 200 trials. c) To evaluate how sensitivity varies with interval between releasing and grasping an object, the y-axis test was repeated with inter-trial intervals of 1, 3, 5, and 7 s. d) To determine how sensitivity varies with type of household object and hand size, the y-axis test was repeated with a telephone, book, hair brush, and television remote and with experimenters with hand sizes ranging from 18.5 to 21.5 cm (tip of middle finger to styloid process of radius). e) To evaluate specificity, the proximity sensor receiver was set > 23 cm away from any transmitters for 24 hours. f) To test robustness to interference from other electronic devices that emit RF waves, the y-axis test was repeated at varying distances from a loud speaker and television set. 2) Movement sensor testing. a) To test how sensitivity varies with distance an object is moved, the movement sensor was attached to the mug. The experimenter moved the mug from one target to another on the table surface parallel to the x-axis of the mug. Two hundred trials each were conducted with the targets 2, 4, 6, 8, 12 and 16 cm apart. The interval between trials was 3 s. b) To test how sensitivity varies with direction of movement, the 12 cm test above was repeated with movements parallel to the y and z axes of the mug. c) To test how sensitivity varies with interval between movements, the12 cm test for movement parallel to the mug s x-axis was repeated with a 2 s inter-trial interval. d) To evaluate specificity, a movement sensor was turned on and left in one spot for 24 hours. 42

51 3) Testing of system. To test the sensitivity and specificity of the entire system, proximity sensor transmitters and movement sensors were attached to two mugs (Mug 1, 2) resting 43 cm apart. The proximity sensor receiver was put on the experimenter s right arm. The experimenter placed his right and left hands on separate targets each > 23 cm from both mugs. When instructed, the experimenter grasped either Mug 1 or 2, moved it to a target 12 cm away with either his Right or Left hand, and returned the hand employed to its starting position. Two hundred trials were conducted, with 5 s between trials. This procedure was repeated with objects set 5 cm apart. The choice of which object to grasp and which arm to employ was determined by a random process on each trial. C. Data Processing and Analysis As noted, the proximity and movement sensor data were stored as text files. A VB.NET software algorithm was developed to process the files offline. The algorithm combined the two files by using the time and ID stamps in these files as keys. Summary variables were then calculated for each test: number of times the experimenter s right arm approached each object (i.e., proximity status transitions from ON to OFF); number of times each object was moved (i.e., movement status transitions from ON to OFF); number of times each object was manipulated by the experimenter s right arm (synchronous transitions from ON to OFF status for the proximity and movement sensors). (Changes in sensor status were deemed synchronous if the transitions in status from each sensor type were within 2 s of each other.) Other summary variables that can be derived are how long each object is moved and manipulated. In addition, total time the arm of interest is used to manipulate objects can be derived by summing across objects. 43

52 IV. RESULTS A. Proximity Sensor Testing Fig. 4 shows that the proximity sensor receiver on the experimenter s right arm detected proximity of an instrumented mug on 100% of trials when the mug was < 21 cm away. Fig. 4. Sensitivity of Proximity Sensor When the mug was 22 cm and 23 cm away, sensitivity fell to 95% and 90%, respectively. When the mug was > 23 cm away, i.e., outside of the intended range, the proximity sensor, appropriately, did not change status. Sensitivity did not vary substantially with angle of approach. Out of 200 approach, grasp, release, and withdraw trials, proximity was detected 202, 198, and 204 times, respectively, for angles parallel to the x, y, and z axes of the mug. Nor did proximity detection vary substantially with interval between trials (1 s = 194, 3 s = 202, 5 s = 198); type of object grasped (mug = 198, telephone = 203; book = 198; hair brush = 194, remote control = 196); or experimenter hand size (18.5 cm = 198, 19.6 cm = 202, 21.5 cm = 204). 44

53 Specificity was supported; no proximity detection signals were recorded when the proximity sensor receiver and transmitter were kept 23 cm apart for 24 hours. In addition, proximity was detected during only 0.4% of inter-trial intervals during the above tests. Operation of a television set and loud speaker interfered with proximity detection; when the proximity sensor receiver and transmitter were within 20 cm of each other but 20 cm from one of these electronic devices the sensor stopped detecting proximity. B. Movement Sensor Testing Fig. 5 shows that when the experimenter moved an instrumented mug 6 cm or more, the movement sensor detected 99% of the movements. For 4 cm movements, detection was 90%. For 2 cm movements, detection was only 57%. Fig. 5. Sensitivity of Movement Sensor Sensitivity did not vary substantially with direction of movement. For 12 cm movements parallel to the x, y, and z axes of the mug, detection was 99%, 99%, and 98%, respectively. Detection was poor when the interval between movements was 2 s. For a 12 cm movement parallel to the x axis of the mug, detection was 99% when the inter-trial interval was 3 s but was only 48% when the inter-trial interval was 2 s. Specificity was supported; no movement was recorded when a movement sensor was 45

54 turned on but kept in one spot for 24 hours. In addition, for tests where the inter-trial interval was 3 s and movement was 4 cm, no movement was detected during the intertrial intervals. C. Testing of System Fig. 6 graphs performance of the SERSMAA system, i.e., joint operation of the proximity and movement sensors when the object to be moved and arm to be employed was randomly selected. Manipulation of the object of interest with the right arm was detected with 100% sensitivity and specificity both when the objects were 43 and 5 cm apart. Fig. 6. Performance of SEARSMAA System V. CONCLUSION The sensitivity and specificity of the SERSMAA prototype under controlled conditions in the laboratory appeared to be adequate for its ultimate purpose, i.e., remotely monitoring everyday arm activity after stroke. When the proximity sensor receiver on the experimenter s right arm drew close ( 21 cm) to an instrumented object, proximity was detected on 97% of trials, regardless of angle of approach, inter-trial interval, type of object, and hand size. When the experimenter s right arm was far ( 23 cm) from an in- 46

55 strumented object, proximity, appropriately, was not signaled. The movement sensor detected 98% of instrumented object movements when they were 6 cm long and 3 s apart, regardless of movement direction. No movement signals were recorded when instrumented objects were at rest. When the object to be manipulated and the arm to be used were randomly selected, the conjoint proximity and movement sensor signals detected handling of the object of interest with the right arm with 100% sensitivity and specificity even when the objects were just 5 cm apart. These results suggest that testing with stroke patients in more natural settings is warranted after some modifications. The size of the sensors needs to be reduced. One approach for doing so would be placing passive RFID tags, which are the size of stickers, on objects, while placing a RF reader with a short range and accelerometer on the arm of interest. A capacity for real-time processing of the system signals would be desirable. The frequency with which household objects in daily life are manipulated by the lessaffected arm when the more-affected arm is within 23 cm of the object, needs to be assessed, as the current system cannot identify which arm has manipulated the object under such conditions. In addition, the frequency of interference from electronic devices such as television sets in everyday environments needs to be assessed. If these issues can be addressed successfully, this technology will be able to provide a much richer objective picture of everyday arm activity after stroke than possible now. Such an advance would permit more accurate measurement of real-world gains after upper-extremity rehabilitation. For this application, the patient s more-affected arm and a representative sample of household objects would be instrumented and the RF reader and PC would be placed in the patient s home for several days before and after rehabilitation. 47

56 Other rehabilitation applications are monitoring compliance with home exercise programs and therapeutic use of activity monitoring records. For example, the SERSMAA output could serve as input for software on the PC controlling a virtual therapist who reinforces patients immediately after they use their more-affected arm to manipulate instrumented objects in their homes. Applications outside of medicine are tracking how often consumers use a company s products (i.e., handle them) and monitoring who handles what on production lines. REFERENCES [1] Lloyd-Jones, D., et al., Heart disease and stroke statistics update: a report from the American Heart Association. Circulation. 121(7): p. e46-e215. [2] Lai, S.M., et al., Persisting consequences of stroke measured by the stroke impact scale. Stroke, : p [3] Nichols-Larsen, D., et al., Factors influencing stroke survivors' quality of life during subacute recovery. Stroke, : p [4] Uswatte, G. and E. Taub, Implications of the learned nonuse formulation for measuring rehabilitation outcomes: lessons from Constraint-Induced Movement therapy. Rehabilitation Psychology, : p [5] Gebruers, N., et al., Monitoring of physical activity after stroke: a systematic review of accelerometry-based measures. Archives of Physical Medicine and Rehabilitation, : p [6] Churko, J.M., et al., Sensor evaluation for tracking upper extremity prosthesis movements in a virtual environment. Conf Proc IEEE Eng Med Biol Soc, : p [7] Parnandi, A., E. Wade, and M. Mataric, Motor function assessment using wearable inertial sensors. Conf Proc IEEE Eng Med Biol Soc. 2010: p [8] Uswatte, G., et al., Ambulatory monitoring of arm movement using accelerometry: an objective measure of upper-extremity rehabilitation in persons with chronic stroke. Archives of Physical Medicine and Rehabilitation, : p

57 [9] Ohashi, K., et al., Comparison of RFID Systems for Tracking Clinical Interventions at the Bedside. AMIA Annu Symp Proc, 2008: p [10] ActiveWave, Inc. Products: conpacttag Datasheet [cited 2011 March 5]; Available from: [11] Kim, D.S., et al., Design of RFID based the Patient Management and Tracking System in hospital. Conf Proc IEEE Eng Med Biol Soc, : p [12] ActiGraph, LLC [cited 2011 March 5]; Available from: [13] Uswatte, G., et al., Objective measurement of functional upper extremity movement using accelerometer recordings transformed with a threshold filter. Stroke, : p [14] Redmond, D.P. and F.W. Hegge, Observations on the design and specification of a wrist-worn human activity monitoring system. Behavior Research Methods, : p

58 CHAPTER 5 SENSOR-ENABLED RFID SYSTEM FOR MONITORING ARM ACTIVITY: RELIABILITY AND VALIDITY by Joydip Barman, Member, IEEE, Gitendra Uswatte, Nilanjan Sarkar, Senior Member, IEEE, Touraj Ghaffari, Brad Sokal, Ezekiel Byrom, Eva Trinh, Michael Brewer, and Christopher Varghese Submitted to IEEE Transactions on Neural Systems and Rehabilitation Engineering Format adapted for dissertation 50

59 Abstract After stroke, capacity to complete tasks in the treatment setting with the moreaffected arm is an unreliable index of actual use of that extremity in daily life. Available objective methods for monitoring real-world arm use rely on placing movement sensors on patients. These methods provide information on amount but not type of arm activity, e.g., functional vs. nonfunctional movement. This paper presents an approach that places sensors on patients and household objects, overcoming this limitation. An accelerometer and the transmitter component of a radio-frequency proximity sensor are attached to objects; the receiver component is attached to the arm of interest. The receiver triggers an on-board radio-frequency identification tag to signal proximity when that arm is within 23 cm of an instrumented object. In benchmark testing, this system detected perfectly which arm was used to move the target object on 200 trials. In a laboratory study with 35 undergraduates, increasing the amount of time target objects were moved with the arm of interest resulted in a corresponding increase in system output (p <.0001). Moreover, measurement error was low ( 2.5%). The results support this system s reliability and Manuscript received March 12, This work was supported by the Department of Psychology at the University of Alabama at Birmingham (UAB), and NIH grant HD Presented in part to the 33rd Annual International Conference of the IEEE Engineering in Medicine and Biology Society. Boston, MA. J. Barman is with the Departments of Biomedical Engineering and Psychology, UAB, Birmingham, AL USA ( joydip@uab.edu) G. Uswatte is with the Departments of Psychology and Physical Therapy, UAB, Birmingham, AL USA ( guswatte@uab.edu) N. Sarkar is with the Department of Mechanical Engineering, Vanderbilt University, Nashville, TN USA ( nilanjan.sarkar@vanderbilt.edu) T. Ghaffari is with ActiveWave Inc., Boca Raton, FL USA ( touraj@activewaveinc.com) B. Sokal is with the Department of Psychology, UAB, Birmingham, AL USA ( bhsokal@uab.edu) E. Byrom is with the Department of Psychology, UAB, Birmingham, AL USA ( zbyrom@uab.edu) E. Trinh is with the Department of Psychology, UAB, Birmingham, AL USA ( evatrinh@uab.edu) M. Brewer is with the Department of Psychology, UAB, Birmingham, AL USA ( mtbrewer@uab.edu) C. Varghese is with the Department of Psychology, UAB, Birmingham, AL USA ( chrisvar@uab.edu) 51

60 validity in individuals with unimpaired movement; testing is now warranted in stroke patients. Index Terms Accelerometer, ambulatory monitoring of stroke survivors, proximity sensor, radio frequency identification or RFID I. INTRODUCTION More than 650,000 individuals survive strokes annually in the United States [1]. Persistent impairment of the arm on the more-affected side of the body afflicts between 55% and 75% of the survivors [2] and is associated with diminished health-related quality of life [3]. Advances in methods to assess and treat the more-affected arm impairment after stroke, therefore, have the potential to improve the quality of life of a large number of people. Well-known models of disability and data indicate that laboratory measures of function poorly index how stroke survivors actually use their more-affected arm in daily life [4]. Therefore, substantial effort has been spent on developing real-world measures of arm function. Most of these tests, however, rely on self-report [4]. Researchers have objectively measured amount of arm activity in the community by placing accelerometers on stroke survivors [5]. These techniques, however, cannot discriminate whether a given arm movement is functional or non-functional and cannot identify what tasks were performed. More complex activity monitors, such as Inertial Measurement Units, hold promise for making such discriminations but to date have been shown only to index quality of 52

61 arm movement after stroke on a standardized motor test in the laboratory [6]-[8]. This paper describes the design and testing of a prototype sensor-enabled radiofrequency identification (RFID) system, which consists of RFID tags paired with proximity and movement sensors for monitoring arm activity. In this system, movement sensors (i.e., accelerometers) are affixed to objects, along with one component of a radiofrequency (RF) proximity sensor. The other component of the RF proximity sensor is connected to an active RFID tag and worn on the arm of interest. Manipulation of instrumented objects with that arm produces synchronous signals from the movement and proximity sensors, permitting tracking of which objects are handled, when handling takes place, and whether handling is by the person and arm of interest. The proposed technology, thus, collects much richer objective data than possible with accelerometers or other physical activity monitors. Data are shared from benchmark testing of the system components (Study 1) and laboratory testing of a prototype system in healthy individuals (Study 2). II. BACKGROUND A. Monitoring of Arm Activity with Accelerometers To overcome the methodological limitations of self-reports, several researchers have employed accelerometers to objectively measure the amount of arm activity in stroke survivors in the community [5]. For example, Uswatte et al. [9] asked stroke survivors with mild-to-moderate impairment of their more-affected arm to wear an accelerometer above each wrist during all waking hours for 2 days before and after upper-extremity physical rehabilitation or a corresponding no-treatment period. The ratio of more- to less- 53

62 affected arm accelerometer recordings was strongly correlated with amount of moreaffected arm use in daily life (r =.74, p <.001). However, since all arm movements produce acceleration readings, this method cannot discriminate whether a given movement is functional or non-functional nor identify what tasks are performed. B. Radio Frequency Identification Systems RFID systems consist of small tags that transmit a unique ID using RF and a RF reader that monitors the status of these tags [10]-[12]. Software on a PC connected to the reader processes the RFID signals. Passive RFID tags transmit their ID when they encounter the reader s radio waves [12], whereas active RFID tags, which are battery powered, transmit their ID independently from as far as 85 m [13]. Applications involve tracking whether tagged objects are within the range of the reader or not. Examples of commercial applications are monitoring when hospital equipment or patients leave designated areas or monitoring changes in inventory of merchandise in a warehouse [14]. However, RFID systems have not been used to remotely monitor upper-extremity activity in stroke survivors or other rehabilitation populations. III. APPARATUS A. Sensor-Enabled RFID System for Monitoring Arm Activity (SERSMAA) Fig. 1 shows the hardware setup of the prototype, and how the movement and proximity sensors operate together when an object is manipulated with the arm of interest. 54

63 Fig. 1. Sketch of SERSMAA Prototype A local area network (LAN) is setup between the PC and a RF reader using an Ethernet 10/100 Mbps switch. The switch enables the reader and PC to communicate reliably over the LAN. A movement sensor (5 cm x 4 cm x 1.7 cm; 37 g) and proximity sensor transmitter (7.8 cm x 3.8 cm x 2 cm; 50 g) are placed on each object. The receiver component of the proximity sensor is connected to an on-board active RFID tag; this assembly (7.4 cm x 6.1 cm x 2.4 cm; 95 g) is attached to the arm of interest. Each movement sensor [15] and the RFID tag [13] possess a unique ID signature. When the arm of interest approaches an instrumented object, the receiver component of the proximity sensor detects the radio transmitter s signals, triggering the RFID tag to broadcast an ON signal along with its ID. When the arm withdraws, the receiver no longer detects a radio signal, triggering the RFID tag to broadcast an OFF signal along with its ID. The RF reader relays the proximity status to the PC, which runs custom VB.NET software that processes the signals and stores the output in a text file. If the object is manipulated, the movement sensor records the changes in its acceleration, and 55

64 stores these values in on-board memory for downloading into a text file that includes the movement sensor ID. Custom software processes the proximity and movement sensor text files offline. Synchronous positive values from the proximity and movement sensors indicate that an instrumented object is being moved by the arm of interest. Moreover, analysis of the proximity status and acceleration values, along with their ID and time stamps, permits tabulation of which objects are moved, when they are moved, for how long, and by which arm. B. Proximity Sensor Fig. 2 and Fig. 3 are block diagrams of the transmitter and receiver components, respectively, of the prototype RF proximity sensor. As described, transmitters are attached to objects, while the receiver is attached to the arm of interest. Oscillator Circuit Low-Frequency Radio Transmitter Fig. 2. Low-Frequency Radio Transmitter Circuit Block Diagram Low-Frequency Radio Receiver Frequency Switch RFID Active Tag Fig. 3. Low-Frequency Radio Receiver Circuit Block Diagram The RF transmitter sends 30 Hz oscillator signals at a fixed low frequency of ~10.7 KHz. The choice of a low-frequency is appropriate for sensing proximity of the receiver and transmitter over short distances of 1 to 23 cm, i.e., for detecting when the arm of interest is adjacent to an instrumented object. The RF receiver is tuned to the same frequency as the transmitter. The receiver output connects to a frequency switch circuit that 56

65 turns ON when it reads the 30 Hz signal and turns OFF in its absence. Because the frequency switch output cannot be readily connected to the ActiveWave RFID tag [13], a jerry-rigged solution is used in this prototype. The frequency switch output, instead of firing the tag directly, connects to an electromagnet, which produces a magnetic field when the frequency switch toggles ON. This magnetic field, in turn, activates an ActiveWave magnetic sensor active RFID tag, which sends a signal to the RF reader indicating a change in sensor status. The power sources for both components are 3 V coin cell batteries. C. Movement Sensor The object movement sensors are ActiGraph GT1M activity monitors. These units employ a biaxial accelerometer, which detects 1 g acceleration with a sensitivity of ±10%. Acceleration is sampled at 60 Hz in each axis. These samples are integrated separately for each axis over a user-specified epoch, which in this case is 1 s, and are stored in 1 Mb flash memory [15]. To generate a single movement status signal for each epoch, OFF is recorded only if the integral values for both axes are 1; else, ON is recorded. The threshold of 1 filters out very small movements that are not likely to be functional [16]. Biaxial accelerometers are adequate for monitoring arm activity because manipulation of objects invariably results in movement components in all 3 axes [17]. A. Methodology IV. STUDY 1: BENCHMARK TESTING Benchmark testing was performed under highly controlled conditions in the labora- 57

66 tory to determine whether the sensitivity and specificity of the SERSMAA prototype was adequate, i.e., 98% and > 99%, respectively [18]. 1) Proximity sensor testing. The proximity sensor transmitter was affixed to the outside of a coffee mug using Velcro. The proximity sensor receiver was attached with an elastic band to the right forearm of the experimenter just above the wrist. The mug was placed on a target at the center of several concentric circles drawn on a tabletop. Two hundred trials of each test were conducted, except for Test 1a, which had 100 trials. The start and end of trials were marked with beeps emitted by custom software on a PC. a) To determine the range of proximity detection, the experimenter moved his hand along the tabletop in 1 cm increments every 5 s starting from a target 24 cm away from the mug and ending 20 cm away. Movement was parallel to the y axis of the mug. Sensor status was logged at each stop. b) To evaluate how sensitivity varies with angle of approach, the experimenter placed his hand on a target > 23 cm from the mug. The experimenter then grasped the mug handle with his right hand, released it, and returned his hand to the target. This movement was conducted parallel to the x, y, and z axes of the mug for separate sets of 200 trials. c) To evaluate how sensitivity varies with interval between releasing and grasping an object, the y-axis test was repeated with inter-trial intervals of 1, 3, 5, and 7 s. d) To determine how sensitivity varies with type of household object and hand size, the y-axis test was repeated with a telephone, book, hair brush, and television remote and with experimenters with hand sizes ranging from 18.5 to 21.5 cm (tip of middle finger to styloid process of radius). e) To evaluate specificity, the proximity sensor receiver was set > 23 cm away from 58

67 any transmitters for 24 hours. f) To test robustness to interference from other electronic devices that emit RF waves, the y-axis test was repeated at varying distances from a loud speaker and television set. 2) Movement sensor testing. a) To test how sensitivity varies with distance an object is moved, the movement sensor was attached to the mug. The experimenter moved the mug from one target to another on the table surface parallel to the x-axis of the mug. Two hundred trials each were conducted with the targets 2, 4, 6, 8, 12 and 16 cm apart. The interval between trials was 3 s. b) To test how sensitivity varies with direction of movement, the 12 cm test above was repeated with movements parallel to the y and z axes of the mug. c) To test how sensitivity varies with interval between movements, the 12 cm test for movement parallel to the mug s x-axis was repeated with a 2 s inter-trial interval. d) To evaluate specificity, a movement sensor was turned on and left in one spot for 24 hours. 3) Testing of system. To test the sensitivity and specificity of the entire system, proximity sensor transmitters and movement sensors were attached to two mugs (Mug 1, 2) set 43 cm apart. The proximity sensor receiver was put on the experimenter s right arm. The experimenter placed his hands on separate targets each >23 cm from both mugs. When signaled, the experimenter grasped either Mug 1 or 2, moved it to a target 12 cm away with either his Right or Left hand, and returned the hand employed to its starting position. Two hundred trials were conducted, with 5 s between trials. This procedure was repeated with objects set 5 cm apart. The choice of which object to grasp and which arm to employ was determined by a random process on each trial. 59

68 4) Data processing and analysis. As noted, the proximity and movement sensor data were stored as text files. A custom-made VB.NET software algorithm combined the files offline by using the time and ID stamps therein as keys. The algorithm then calculated the number of following events for each test and object of interest: experimenter s right arm approached object (i.e., proximity status transitions from ON to OFF); object was moved (i.e., movement status transitions from ON to OFF); object was manipulated by the experimenters right arm (synchronous transitions from ON to OFF status for the proximity and movement sensors). Changes in sensor status were deemed synchronous if the transitions in status from each sensor type were 2 s apart. B. Results 1) Proximity sensor testing. Fig. 4 shows how the sensitivity of the proximity sensor receiver varied with distance from an instrumented mug. When the mug was > 23 cm away, i.e., outside of the intended range, the proximity sensor, appropriately, did not change status. Fig. 4. Sensitivity of Proximity Sensor 60

69 Sensitivity did not vary substantially with angle of approach. Out of 200 approach, grasp, release, and withdraw trials, proximity was detected 202, 198, and 204 times, respectively, for angles parallel to the x, y, and z axes of the mug. Nor did proximity detection vary substantially with interval between trials (1 s = 194, 3 s = 202, 5 s = 198); type of object grasped (mug = 198, telephone = 203; book = 198; hair brush = 194, remote control = 196); or experimenter hand size (18.5 cm = 198, 19.6 cm = 202, 21.5 cm = 204). Specificity was supported; no proximity detection signals were recorded when the proximity sensor receiver and transmitter were kept 23 cm apart for 24 hours. In addition, proximity was detected during only 0.4% of inter-trial intervals during the above tests. Operation of a television and loudspeaker interfered with proximity detection; when the proximity sensor receiver and transmitter were within 20 cm of each other but 20 cm from these devices the sensor stopped detecting proximity. 2) Movement sensor testing. Fig. 5 shows how the sensitivity of the movement sensor varied with distance the instrumented object was moved. Fig. 5. Sensitivity of Movement Sensor 61

70 Sensitivity did not vary substantially with direction of movement. For 12 cm movements parallel to the x, y, and z axes of the mug, detection was 99%, 99%, and 98%, respectively. Detection was poor when the interval between movements was 2 s. For a 12 cm movement parallel to the x axis of the mug, detection was 99% when the inter-trial interval was 3 s but only 48% when it was 2 s. Specificity was supported; no movement was recorded when a movement sensor was turned on but kept in one spot for 24 hours. In addition, for tests where the inter-trial interval was 3 s and movement was 4 cm, no movement was detected during the intertrial intervals. 3) Testing of system. On the test in which the object to be moved and the arm to be employed were randomly selected, manipulation of the object of interest with the right arm was detected by the system with 100% sensitivity and specificity both when the objects were 43 and 5 cm apart. V. STUDY 2: RELIABILITY AND VALIDITY A. Methodology To evaluate whether SERSMAA reliably and validly measures manipulation of household objects, 3 tests were conducted in a 2.7 m x 1.5 m room in our laboratory. Participants were recruited from the introductory psychology subject pool at our university; 35 students with unimpaired upper-extremity movement were enrolled (27 women; median age = 19 yr, range = 17-46; 8 men; median age = 19 yr, range = 17-24). For all 3 tests, the proximity sensor receiver was fastened just above the wrist of the right arm of participants with an elastic band. A proximity sensor transmitter and an accelerometer 62

71 were affixed with Velcro to five objects (Fig. 6). For Tests 1 and 2, the objects were handled in a fixed order. For Test 3, the objects to be handled by each participant in each experimental condition (Low, Medium, High) were selected randomly. For all 3 tests, order of the conditions was counterbalanced across participants, with breaks given between conditions of 10, 20, and 30 s in Tests 1, 2, and 3, respectively. 1) Test 1. The five household objects were placed 28 cm apart in a line on a tabletop. Fig. 6 gives the arrangement of the objects and starting and target positions. Target Location cm Starting Location 14 cm Mug Book Hairbrush Telephone Remote Fig. 6. Setup of Objects for Test 1 and 2 Participants moved each object from its starting to target position and back to a 40 bpm rhythm for 6 out of 18 s (Low), 12 out of 18 s (Medium), and 18 out of 18 s (High). The Medium condition was repeated before proceeding to the next object. 2) Test 2. The procedure was similar to Test 1 except that instead of letting the object rest for 12 and 6 s in the Low and Medium conditions, participants moved it with their left hand. In addition, 5 s were given to switch from the right to left hand in these conditions. 3) Test 3. All 5 objects were placed at one end of the table. Participants, using their right arm, moved 1 object (Low), 3 objects (Medium), and 5 objects (High) to the opposite end, which was 150 cm away. As for Test 1 and 2, the Medium condition was repeated. So that the movements would be more naturalistic than in Test 1 and 2, the par- 63

72 ticipants behavior was not regulated in other way. 4) Data processing and analysis. Data processing was similar to Study 1. The summary variable (SV) calculated was time each object was handled with the right arm during each experimental condition. To evaluate reliability, measurement error was quantified for each target object and condition in Test 1 and 2 by expressing the difference between the actual duration of object handling and SERSMAA SV value as a percentage of the actual duration. In addition, the similarity of the SV values from the first and second iteration of the Medium conditions in Tests 1-3 was evaluated by calculating the mean difference in SV values between the first and second iteration and corresponding twotailed, 95% confidence interval (CI). Correlational methods for evaluating test-retest reliability were not appropriate because successful implementation of the experimental manipulation, i.e., amount of time target objects were handled with the right arm, and low measurement error made the range of SV values for the first and second iteration of the Medium condition very small. Construct validity was evaluated by testing whether increasing amount of time that objects were handled with the right arm (Test 1 & 2) or number of objects that were handled with the right arm (Test 3) produced a corresponding increase in SERSMAA SV values. For this purpose, separate analysis of variance (ANOVA) models were specified for Test 1 and 2 data with two repeated measures factors: Amount of Handling (Low, Medium, High) and Object (1-5). The ANOVA model for Test 3 data had only one repeated measures factor: Number of Objects (Low, Medium, High). The Huynh-Feldt correction was applied to the p-values reported to account for any violations of the sphericity assumption. Data from one participant were missing because of technical issues with the system during data collection. 64

73 B. Results Error for measuring amount of time objects were handled with the right arm was only 2.5% (Test 1: mean error = 3.4%, SD = 1.3; Test 2: mean error = 1.6%, SD = 1.0). SERSMAA SV values from the first and second iteration of the Medium condition were highly similar (Test 1: mean difference = 0.2 s, CI = ; Test 2: mean difference = 0.05 s, CI = ; Test 3: mean difference = 0.1 s, CI = ). In addition, when only one object was handled, SERSMAA erroneously reported handling of more than one less than 1% of the time (Test 1: mean = 0.5%, SD = 0.7; Test 2: mean = 0.8%, SD = 0.6). In Test 3, in which the objects to be handled in each condition were assigned randomly, SERSMAA correctly reported the objects handled on all trials. Increasing the amount of time the target object was handled (Test 1 & 2) or the number of target objects handled (Test 3) with the right arm resulted in a corresponding increase in system output (Test 1: F[2, 324] = 1997, p <.0001, Huynh-Feldt ε = 0.87; Test 2: F[2, 328] = 4477, p <.0001, ε = 0.97; Test 3: F[2, 66] = 445, p <.001, ε = 1.03 ). Table I reports mean seconds target objects were handled with the right arm in each experimental condition for Tests 1-3. Data were collapsed across the second repeated measures factor, i.e., Object, for Test 1 and 2 because there was no main effect for Object or Amount of Time x Object interaction. 65

74 TABLE 1. Time (s) Objects Were Manipulated With the Right Arm per SERSMAA Output (N = 34) Test 1 Test 2 Test 3 Low Medium 1 High Medium ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) Note. Values are mean, confidence interval (lower bound upper bound). VI. CONCLUSION AND DISCUSSION The first application proposed for the SERSMAA system is remotely monitoring everyday arm activity after stroke. This system, unlike existing methods for ambulatory monitoring of arm activity, captures information about the movement and proximity of household objects. It does so by placing movement and proximity sensors on the arm and objects of interest. The results of the benchmark testing (Study 1) and laboratory study with individuals with unimpaired movement (Study 2) suggest that the SERSMAA prototype embodies a promising approach. The benchmark testing supported the sensitivity and specificity of the SERSMAA prototype for detecting proximity of the arm of interest to a household object, movement of the object, and handling of the object. When the proximity sensor receiver on the experimenter s right arm drew close ( 21 cm) to an instrumented object, proximity was detected on 97% of trials, regardless of angle of approach, inter-trial interval, type of object, and hand size. When the experimenter s right arm was far ( 23 cm) from an instrumented object, proximity, appropriately, was not signaled. The movement sensor de- 66

75 tected 98% of instrumented object movements when they were 6 cm long and 3 s apart, regardless of movement direction. No movement signals were recorded when instrumented objects were at rest. When the object to be manipulated and the arm to be used were randomly selected, the conjoint proximity and movement sensor signals detected handling of the object of interest with the right arm with 100% sensitivity and specificity even when the objects were just 5 cm apart. The laboratory study supported the reliability and construct validity of the SERSMAA prototype for measuring amount of time household objects were handled with the arm of interest. Amount of time objects were handled with the participants right arm was measured with low error ( 2.5%) and the SERSMAA output was very similar for two iterations of the same condition. The mean difference in output values between the first and second iteration was < 1% of the time objects were handled. Increasing the amount of time household objects were handled, as predicted, produced a corresponding increase in SERSMAA output (p s <.0001). In addition, the prototype discriminated objects that were handled from those that were not with > 99% accuracy. These results suggest that future studies with stroke survivors in more natural settings are warranted. However, two modifications to the system are necessary. First, the size of the sensors needs to be reduced. An approach for doing so would be placing passive RFID tags [10]-[12], which are the size of a large stamp, on objects, while placing a RF reader with a short range and accelerometer on the arm of interest. Second, a capacity for real-time processing of the system signals would be desirable. Two other limitations of the prototype need further study. The frequency with which household objects in daily life are manipulated by the less-affected arm when the more-affected arm is within 23 cm 67

76 of the object, needs to be assessed, as the current system cannot identify which arm has manipulated the object under such conditions. In addition, the frequency of interference from electronic devices such as television sets in everyday environments needs to be assessed. If these issues can be addressed successfully, this technology will be able to provide a much richer objective picture of everyday arm activity after stroke than possible now. Such an advance would permit more accurate measurement of real-world gains after upper-extremity rehabilitation. For this application, the patient s more-affected arm and a representative sample of household objects would be instrumented, and the RF reader and PC would be placed in the patient s home for several days pre- and post-therapy. Other rehabilitation applications are monitoring compliance with home exercise programs and therapeutic use of activity monitoring records. For example, the SERSMAA output could serve as input for software on the PC that controls a virtual therapist who reinforces patients immediately after they use their more-affected arm to manipulate instrumented objects in their homes. Business applications include tracking how often customers use a company s products (i.e., handle them) and monitoring which employees handle what on production lines. REFERENCES [1] D. Lloyd-Jones, et al., "Heart disease and stroke statistics update: a report from the American Heart Association", Circulation, vol. 121, pp. e46-e215, Feb 23, [2] S.M. Lai, S. Studenski, P.W. Duncan, and S. Perera, "Persisting consequences of stroke measured by the stroke impact scale", Stroke, vol. 33, pp ,

77 [3] D. Nichols-Larsen, P.C. Clark, A. Zeringue, A. Greenspan, and S. Blanton, "Factors influencing stroke survivors' quality of life during subacute recovery", Stroke, vol. 36, pp , [4] G. Uswatte and E. Taub, "Implications of the learned nonuse formulation for measuring rehabilitation outcomes: lessons from Constraint-Induced Movement therapy", Rehabil Psychol, vol. 50, pp , [5] N. Gebruers, C. Vanroy, S. Truijen, S. Engelborghs, and P.P. De Deyn, "Monitoring of physical activity after stroke: a systematic review of accelerometry-based measures", Arch Phys Med Rehabil, vol. 91, pp , [6] M.J. Johnson, Y. Shakya, E. Strachota, and S.I. Ahamed, "Low-cost monitoring of patients during unsupervised robot/computer assisted motivating stroke rehabilitation", Biomed Tech (Berl), vol. 56, pp. 5-9, Feb, [7] J.M. Churko, A. Mehr, A. Linassi, and A. Dinh, "Sensor evaluation for tracking upper extremity prosthesis movements in a virtual environment", in Conf Proc IEEE Eng Med Biol Soc, Minneapolis, MN, 2009, pp [8] A. Parnandi, E. Wade, and M. Mataric, "Motor function assessment using wearable inertial sensors", in Conf Proc IEEE Eng Med Biol Soc, Buenos Aires, Argentina, 2010, pp [9] G. Uswatte, et al., "Ambulatory monitoring of arm movement using accelerometry: an objective measure of upper-extremity rehabilitation in persons with chronic stroke", Arch Phys Med Rehabil, vol. 86, pp , [10] J. Kabachinski, "An introduction to RFID", Biomedical Instrumentation & Technology, vol. 39, pp , [11] K. Ohashi, S. Ota, H. Tanaka, and L. Ohno-Machado, "Comparison of RFID Systems for Tracking Clinical Interventions at the Bedside", in AMIA Annu Symp Proc, Washington, DC, 2008, pp [12] P. Kumar, H.W. Reinitz, J. Simunovic, K.P. Sandeep, and P.D. Franzon, "Overview of RFID technology and its applications in the food industry", J Food Sci, vol. 74, pp. R101-6, Oct, [13] ActiveWave, Inc., (2011, March 5), Products: conpacttag Datasheet. [Online] Available: [14] D.S. Kim, J. Kim, S.H. Kim, and S.K. Yoo, "Design of RFID based the Patient Management and Tracking System in hospital", in Conf Proc IEEE Eng Med Biol Soc, Vancouver, Canada, 2008, pp

78 [15] ActiGraph, LLC., (2011, March 5) Available: [16] G. Uswatte, et al., "Objective measurement of functional upper extremity movement using accelerometer recordings transformed with a threshold filter", Stroke, vol. 31, pp , [17] D.P. Redmond and F.W. Hegge, "Observations on the design and specification of a wrist-worn human activity monitoring system", Behavior Research Methods, vol. 17, pp , [18] J. Barman, G. Uswatte, N. Sarkar, T. Ghaffari, and B. Sokal, "Sensor-Enabled RFID System for Monitoring Arm Activity in Daily Life", in Conf Proc IEEE Eng Med Biol Soc, Boston, MA, 2011, pp Joydip Barman (M 11) received the M.S. in electrical engineering from University of Alabama at Birmingham (UAB), Birmingham, AL, USA, in 2004, and currently pursuing Ph.D. in biomedical engineering at UAB, Birmingham, AL, USA. He is currently a Research Engineer with the Constraint-Induced Therapy Research Group in the Department of Psychology at UAB. Gitendra Uswatte received the B.A. in economics from Princeton University, Princeton, NJ, USA, in 1991, and the Ph.D. in clinical health psychology from UAB, Birmingham, AL, USA, in He is currently an Associate Professor of Psychology and Physical Therapy at UAB. Nilanjan Sarkar received the B.E. in mechanical engineering from University of Calcutta, Kolkata, India, in 1985, and the Ph.D. in mechanical engineering from University of Pennsylvania, Philadelphia, PA, USA, in He is currently a Professor of Mechanical and Computer Engineering at Vanderbilt University and a Senior Member of IEEE. Touraj Ghaffari received the B.S in electrical engineering from Kansas University, Lawrence, KS, USA, in 1976, and the M.S. in electrical engineering from University of Missouri, Columbia, MO, USA, in He is currently President and CEO/Founder of ActiveWave, Inc. Brad Sokal received the B.A. in english and art history from Oberlin College, Oberlin, OH, USA, in 1998, and currently pursuing the Ph.D. in clinical psychology from UAB, Birmingham, AL. Ezekiel Byrom is a senior in the B.S. degree in psychology at UAB, Birmingham, AL, USA. Eva Trinh is a senior in the B.S. degree in psychology at UAB, Birmingham, AL, USA. 70

79 Michael Brewer is a junior in the B.S. degree in psychology at UAB, Birmingham, AL, USA. Christopher Varghese received the B.S. in psychology from UAB, Birmingham, AL, USA, in 2011, and currently pursuing the M.S. in psychology from Marshall University, WV, USA. 71

80 CHAPTER 6 CONCLUSION AND DISCUSSION Design Considerations The movement sensor considered was an off-the-shelf product supplied by Acti- Graph. Since the specifications for the ActiGraph GT1M activity monitor satisfied the requirements for measuring arm activity, no further investigation was carried out on another technology. The size of this sensor is a limitation and needs further research to not only reduce the overall size but also integrate the sensor with the RFID system, so as to enable the RFID tracking software to simultaneously track the movement and proximity sensors and provide a real-time feedback of the movement of the object with respect to the tagged arm. The current system does this deduction after post processing of the sensor data files. Benchmark testing of the movement sensor (Chapter 4) showed that the sensor detected 98% of instrumented object movements when they were 6 cm long and 3 s apart, regardless of movement direction. No movement signals were recorded when instrumented objects were at rest. The hardware design of the prototype proximity sensor (Chapter 3 Aim 1) proved to satisfy the design constraints set forth in consultation with the clinicians. The hardware design process for the proximity sensor included considering present technologies like infrared and magnetic proximity sensors. Wireless technology for the prototype proximity sensor was considered since it addresses the issues of the other proximity sensor technologies. The prototype low-frequency radio proximity sensor was designed in two parts; a transmitter component to be attached to the object and a receiver component to be 72

81 worn on the more-impaired arm. The detection range for the proximity sensor was set at < 23 cm from the object after considering several household objects and also considering the mounting location for the transmitter on the object and the respective distance from the transmitter to the receiver on the wrist. Benchmark testing (Chapter 4) showed that when the proximity sensor receiver on the experimenter s right arm drew close ( 21 cm) to an instrumented object, proximity was detected on 97% of trials, regardless of angle of approach, inter-trial interval, type of object, and hand size. When the experimenter s right arm was far ( 23 cm) from an instrumented object, proximity, appropriately, was not signaled. The developed prototype proximity sensor met and exceeded the performance standards and design constraints. Custom RFID system software was developed in VB.NET to assist the user in tracking the proximity sensor, process data files offline for computing summary variables and provide an assessment of the tracking session. The software was developed in two parts based on the design constraints set forth in Chapter 3 Aim 2; the RFID tracking software and the RFID data processing software. As the name suggests, each of these software had a very specific task to accomplish. The tracking software helped the user with establishing a LAN communication channel for the data packets to be acquired from the RF reader via the network switch, and save the proximity status data in a text file. The data processing software enabled the user to process these huge data files obtained from the tracking software and the text file created after downloading the GT1M activity monitor. The data processing software reduced the manual processing time of these text files from several hours to a matter of few minutes. The accelerometer text file (GT1M file) needed three step transformations before it can be matched with the proximity text file using the com- 73

82 mon time stamp. The processed combined data file was then used to compute the summary variables such as the total time in movement of the tagged arm and total time in movement of each of the tagged objects. The summary variables were further analyzed by SAS 9.1 statistical software. Benchmark testing of the RFID system (Chapter 4) showed that when the object to be manipulated and the arm to be used were randomly selected, the conjoint proximity and movement sensor signals detected handling of the object of interest with the right arm with 100% sensitivity and specificity even when the objects were just 5 cm apart. To conclude, the custom RFID system software satisfied and exceeded the performance standards and design constraints. The laboratory study with normal college students (Chapter 3 Aim 4 and Chapter 5) supported the reliability and construct validity of the prototype RFID system for measuring amount of time household objects were handled with the arm of interest. This study helps answer the research questions set forth in Chapter 3 Aim 4 for evaluating the prototype RFID system. Amount of time objects were handled with the participants right arm was measured with low error ( 2.5%) and the system output was very similar for two iterations of the same condition. The mean difference in output values between the first and second iteration was < 1% of the time objects were handled. Increasing the amount of time household objects were handled, as predicted, produced a corresponding increase in system output (p s <.0001). In addition, the prototype discriminated objects that were handled from those that were not with > 99% accuracy. Study Limitations The prototype RFID system was designed to capture functional task-related arm movements with respect to a sample of commonly used household objects. The ability of 74

83 the system to track non-task related functional arm movements was not evaluated, since this was not a design objective. As noted, rehabilitation emphasizes training to improve how well and how often patients complete everyday tasks. Thus, measurement of nontask related functional movement is of less interest than task-related functional movements. Another limitation was that the current design did not permit tagging of all objects in the home environment. At the current state of the system, only a limited number of objects could be tagged due the size of the sensors. Further research and development is required to reduce the size of the sensors, which will allow tagging of more household objects. Thus, the sample of objects will better represent the population the tasks commonly performed in a home environment. This expanded sample of objects will help better evaluate arm use on a wider range of tasks related activities. The prototype RFID system was able to capture all functional movements associated with the tagged arm with respect to the tagged household objects. Further testing is required to evaluate whether the system has specificity for detecting functional movements, i.e., whether it does not give false positives in response to non-functional arm movements. A pilot study with twenty-five stroke patients involved comparing the RFID system output to observational behavioral coding of the video tapes for tasks performed by these stoke patients in the laboratory. In this study the subjects were involved in both object related movement, and non-object related movement of the target arm. This study demonstrated that the prototype RFID system in fact captures functional use of the arm with both sensitivity and specificity. The current size of the sensors limits the use of this system in the real life environment, and therefore further research is required to reduce the size of the sensors. An ap- 75

84 proach for doing so would be placing passive RFID tags [14], which are the size of a large stamp, on objects, while placing a RF reader with a short range and accelerometer on the arm of interest. A capacity for real-time processing of the system signals would be desirable in the next generation of this system. Two other limitations of the prototype need further study. The frequency with which household objects in daily life are manipulated by the less-affected arm when the more-affected arm is within 23 cm of the object, needs to be assessed, as the current system cannot identify which arm has manipulated the object under such conditions. In addition, the frequency of interference from electronic devices such as television sets in everyday environments needs to be assessed. Implications and Future Studies If the limitations of the current system can be addressed successfully, the prototype RFID system can be implemented to remotely monitor everyday arm activity after stroke. This system, unlike existing methods for ambulatory monitoring of arm activity, captures information about the movement and proximity of household objects with respect to the tagged arm. This technology will be able to provide a much richer objective picture of everyday arm activity after stroke than possible now. Such an advance would not only permit more accurate measurement of real-world gains after upper-extremity rehabilitation but also enable in better diagnosing deficits in use of the more-impaired arm than possible now, and thus help improve treatment planning. For this application, the patient s more-affected arm and a representative sample of household objects would be instrumented, and the RF reader and PC would be placed in the patient s home for several days pre- and post-therapy. Other rehabilitation applications are monitoring compliance 76

85 with home exercise programs and therapeutic use of activity monitoring records. For example, the prototype RFID system output could serve as input for software on the PC that controls a virtual therapist who reinforces patients immediately after they use their moreaffected arm to manipulate instrumented objects in their homes. Other possible areas of heath care where objective assessment of patient activity is important and the proposed technology might be valuable are geriatrics, obesity, and compliance with exercise programs. Business applications include tracking how often customers use a company s products (i.e., handle them) and monitoring which employees handle what on production lines. 77

86 REFERENCES [1] E. Taub, G. Uswatte, VW. Mark, DM. Morris, "The learned nonuse phenomenon: implications for rehabilitation", Eura Medicophys, 42(3), pp , Sep [2] AS. Adams, SB. Soumerai, J. Lomas, D. Ross-Degnan, "Evidence of self-report bias in assessing adherence to guidelines", Int J Qual Health Care,11(3), pp , Jun [3] CL. Kimberlin, AG. Winterstein, "Validity and reliability of measurement instruments used in research", Am J Health Syst Pharm, 65(23), pp , Dec [4] HP. Lacey, A. Fagerlin, G. Loewenstein, DM. Smith, J. Riis, PA. Ubel, "Are they really that happy? Exploring scale recalibration in estimates of well-being", Health Psychol, 27(6), pp , Nov [5] G. Lewis, "Observer bias in the assessment of anxiety and depression", Soc Psychiatry Psychiatr Epidemiol, 26(6), pp , Dec [6] JM. Lyness, C. Cox, J. Curry, Y. Conwell, DA. King, ED. Caine, "Older age and the underreporting of depressive symptoms", J Am Geriatr Soc, 43(3), pp , Mar [7] H. Rosenberg, "Clinical and laboratory assessment of the subjective experience of drug craving", Clin Psychol Rev, Jun [8] G. Uswatte, WH. Miltner, B. Foo, M. Varma, S. Moran, E. Taub, "Objective measurement of functional upper-extremity movement using accelerometer recordings transformed with a threshold filter", Stroke, 31(3), pp , Mar [9] EA. Fry, LA. Lenert, "MASCAL: RFID tracking of patients, staff and equipment to enhance hospital response to mass casualty events", AMIA Annu Symp Proc, pp , [10] KH. Gamble, "Tracking technologies: real-time tracking technology may promise futuristic functionality, but many CIOs are using it today to save money", Healthc Inform, 25(14), pp , Feb [11] E. Iadanza, F. Dori, R. Miniati, R. Bonaiuti, "Patients tracking and identifying inside hospital: a multilayer method to plan an RFId solution", Conf Proc IEEE Eng Med Biol Soc, pp , [12] DS. Kim, J. Kim, SH. Kim, SK. Yoo, "Design of RFID based the Patient Manage- 78

87 ment and Tracking System in hospital", Conf Proc IEEE Eng Med Biol Soc, pp , [13] CC. Lin, PY. Lin, PK. Lu, GY. Hsieh, WL. Lee, RG. Lee, "A healthcare integration system for disease assessment and safety monitoring of dementia patients", IEEE Trans Inf Technol Biomed, 12(5), pp , Sep [14] K. Ohashi, S. Ota, H. Tanaka, L. Ohno-Machado, "Comparison of RFID Systems for Tracking Clinical Interventions at the Bedside", AMIA Annu Symp Proc, pp , [15] E. Taub, "Movement in nonhuman primates deprived of somatosensory feedback, Exerc Sport Sci Rev, 4, pp , [16] E. Taub, JE. Crago, G. Uswatte, "Constraint-induced movement therapy: A new approach to treatment in physical rehabilitation", Rehabilitation Psychology, 43(2), pp , [17] E. Taub, et al, "Technique to improve chronic motor deficit after stroke", Arch Phys Med Rehabil, 74(4), pp , Apr [18] G. Uswatte, E. Taub, "Implications of the learned nonuse formulation for measuring rehabilitation outcomes: lessons from Constraint-Induced Movement Therapy", Rehabilitation Psychology, 50, pp , [19] MM. Ouellette, et al, "High-intensity resistance training improves muscle strength, self-reported function, and disability in long-term stroke survivors", Stroke, 35(6), pp , Jun [20] G. Uswatte, E. Taub, D. Morris, K. Light, PA. Thompson, "The Motor Activity Log-28: assessing daily use of the hemiparetic arm after stroke", Neurology, 67(7), pp , Oct [21] G. Uswatte, E. Taub, D. Morris, M. Vignolo, K. McCulloch, "Reliability and validity of the upper-extremity Motor Activity Log-14 for measuring real-world arm use", Stroke, 36(11), pp , Nov [22] ActiveWave, Inc., (2011, March 5), Products: conpacttag Datasheet. [Online] Available: [23] P. Naveen, (August 10, 2009), "Infrared beam barrier/ proximity sensor", [Online], Available: [24] I. Kamal, (August 10, 2009), "Infra-Red Proximity Sensor (I)", [Online], Available: 79

88 APPENDIX 1 IRB APPROVAL LETTER 80

89 APPENDIX 2 ACTIVE RFID TAG SPECIFICATIONS User Memory Multi-Tag Read Capability Transmit Frequency Receive Frequency Read range Power Battery Life Dimensions Weight Case Material Temperature Tag options Kbits Yes 916 MHz, 927 MHz, or 868 MHz 433 MHz Receive 30m (100 feet) Transmit 45m (150 feet) 3V Lithium-ion replaceable watch battery 1-3 years depending on use (tag has low battery detection) 59.9 mm x 30.5 mm x 10.2 mm (2.4 in x 1.2 in x 0.4 in) 14 grams (0.5 oz) ABS (Acrylonitrile Butadiene Styrene) Operating -35C to +50C (-31F to +122F) Storage Tamper LED Buzzer Memory -40C to +85C (-40F to +185F) Alarms if tag removed from optional bracket Blinks when called Beeps when called 0-128Kbit memory sizes available in 2x increments 81

90 APPENDIX 3 RFID READER SPECIFICATIONS Functionality Reads and writes RFID tags Multi-Tag Read Capability Yes Transmit Frequency to Tag 433 MHz Receive Frequency from Tag 916 MHz, 927 MHz, or 868 MHz Range 30m (100 feet) to tag 85m (280 feet) from tag RS232 Host Communications Ethernet WLAN (optional) Power 12Vdc, 1.5A Dimensions Weight Case Material Temperature Indicators Connectors without antennas Baud 10/100 Mbps 2.4 GHz, 5.2 GHz 150 mm x 85 mm x 27 mm (5.9 in x 3.3 in x 1.1 in) 150 mm x 85 mm x 167 mm (5.9 in x with antennas 3.3 in x 6.6 in) 680 grams (1.5 lbs) Impact resistant polystyrene with UL94-HB flammability rating Operating -35C to +50C (-31F to +122F) Storage RF LED -40C to +85C (-40F to +185F) On while receiving packet from tag. HOST LED On while sending validated tag packet to Host. ACCESS LED On while transmitting packet to tag. POWER LED On when Reader is powered. Power 12Vdc, 1.5A Ethernet RJ-45 female to Host Motion Detector RJ-11 male Host Comm. Input Output Same RJ-11 male to Host (DB9 female to Host optional) Two contact sense inputs Two isolated dry contact relay outputs 82

91 APPENDIX 4 ACTIGRAPH GT1M ACCELEROMETER SPECIFICATIONS 83

92 84