540 likes | 789 Views
Medical Embedded Device for Individualized Care (MEDIC). Winston Wu. August 20, 2004. Outline. Background Application of ENS to Medical Informatics New Approach Hierarchical diagnostics System Design Requirements Sensor systems Hardware and software architecture Implementation
E N D
Medical Embedded Device for Individualized Care (MEDIC) Winston Wu August 20, 2004
Outline • Background • Application of ENS to Medical Informatics • New Approach • Hierarchical diagnostics • System Design • Requirements • Sensor systems • Hardware and software architecture • Implementation • Experimental results • On-node diagnostics • Next steps and opportunities
UCLA Medical Informatics Group • UCLA Medical Informatics in the School of Medicine • Lead by Dr. Hooshang Kangarloo • Co-Director of a new Center collaboration with HSSEAS • Critical guidance for development of new ENS applications • Broad range of research programs • Diagnostics based on multimedia information processing • Imaging (diverse, MRI, PET) • Clinical information • Natural language • Bayesian network automated reasoning methods • Open source tools for medical informatics • Education programs • National Library of Medicine • http://www.mii.ucla.edu/
Background: Cardiovascular Disease • Prevalence—64.4 million (22.6%) • 25.3 million age 65 or older • 7.8 million suffered heart attack • 5 million congestive heart failure • 4.8 million suffered stroke • Mortality—930,000 (38.5% of all deaths) • Extended clinical treatment • 6.2 million patients • Health care cost—$370 billion • Adequate health care resources are not available for all patients! (source) Data for Americans in 2001 by American Heart Association (2004)
Medical Informatics Vision • Develop information technology: • Address primary challenges in disease detection • Early detection of disease increases success of treatment • Early detection of stroke reduces resulting disability • Improve likelihood of positive outcome • Reduce healthcare costs • Particular focus on cardiovascular disease • Enable a transition from subjective to objective diagnostics • Based on diverse data sources • Need information on patients in the real world
Medical Informatics: Need for Continuous Patient Monitoring • Acquire data from patients in the real world • Currently available wearable systems • Simply acquire and forward information to centralized server systems • Some event detection being developed • Limitations • Practical wearable sensor choices are limited • Diagnostics capability is also, therefore, limited • (Often, many diverse sensor and instrument data sources are required for evidence to support a diagnostic hypothesis)
Approach for Individualized Diagnosis • Recognize that limited wearable sensors are not sufficient to form a high certainty hypothesis regarding patient diagnosis • New MEDIC approach exploits fusion of: • Local (limited) sensor population • Local, networked instruments • Local information processing • Evidence from centralized data resources • Data from direct patient query • Responses to limited and standard verbal queries • SF36 Standard • Require local reasoning by wearable to identify high utility actions • Leverage considerable experience and technology in UCLA Medical Informatics
Overview • Background • Objective diagnostics motivation • Approach • Hierarchical diagnostics • System Design • Requirements • Sensor systems • Hardware and software architecture • Implementation • Experimental results • On-node diagnostics • Next steps
Individuals with wearable ENS Local Sensor Information Processing Centralized Medical Informatics Sensor Records, Events Modular Algorithms Centralized Medical Informatics Internet System Architecture
Modular Sensors instrument Modular Wearable Platform instrument Centralized Medical Informatics Sensor Records and Events Modular Algorithms Internet System Architecture Focusing on One Individual
Overview • Background • Objective diagnostics motivation • Approach • Hierarchical diagnostics • System Design • Requirements • Sensor systems • Hardware and software architecture • Implementation • Experimental results • On-node diagnostics • Next steps
Platform Requirements • Information Acquisition • Local wearable sensors • Electrocardiogram (ECG), Electromyogram (EMG), respiration, skin temperature, CO2 • Pulse rate, oxygen saturation (SpO2), motion tracking (gait) • Remote sensors • Blood pressure, weight scale • User interfaces • Audio / speech • Sensor Interfaces • Local wearable sensors • analog, serial, USB, infrared, etc. • Remote sensors • 802.11, Bluetooth, ZigBee
Platform Requirements (cont) • Processing requirements • Continuously vigilant event detection • Sufficient processing capability for hosting complex algorithms (sensor signal processing, sensor data fusion). • Communication requirements • Use existing standards • Wireless (802.11, 802.15, ZigBee, etc.) • Security (SSL) • Other standard protocols (HTTP) • Energy requirements • Applications permit rechargeable operations • 24-48 hours
Hardware Platform • Processor / Preprocessor Approach • Apply low power preprocessor for sensor sampling and platform management • Operate preprocessor continuously • Apply 32-bit embedded processor hosting Linux OS for application support • Operate processor at low duty cycle (~ 1 %) • Energy benefits • Reduce energy usage without sacrificing vigilance • Incorporate Emstar into platform host – preprocessor interface • Enable uniform sensor interfaces and platform management
Sensor Sensor Sensor Centralized Medical Informatics Access Point Instrument Wearable Internet Sensor Data Acquisition & Event Detection Feedback -Event detection thresholds -Fusion policy -Patient-specific data System Design
Standalone Instruments • Support RS-232C serial interface • Connected to a embedded platform with Linux OS to form standalone sensor instrument • Embedded platform supports wireless services and authentication protocols • User commands measurement and data is automatically transferred to the wearable • Examples • OMRON HEM-747IC Automatic Digital Blood Pressure • A&D UC-321PS Weight Scale
Oximeter Sensor • Simple non-invasive method of monitoring the percentage of hemoglobin that is oxygen saturated • Based on optical spectroscopy of light absorbed by oxygen molecules in blood • Supports RS-232 serial interface • Connected to the preprocessor • Example • NONIN Xpod oximeter
z y x Motion Sensing X Y Z Analog Data To Microcontroller Motion Sensors for Gait Analysis • Low power (2mW) • Senses linear acceleration • Example • ADXL311EB from Analog Devices • Use 3 for motion in 3-D • Sensitivity (167mV/g)
Immediate Goals • Initiate development of sensor systems and algorithms leading to capability for stroke and other cardiovascular disease detection • Sensing requirements include • Blood oxygen • Pulse • Gait • Weight • Speech queries • Blood oxygen and pulse information available now • Posture and gait identification current priority • Identify posture, limp, run, walk, stand, etc.
Standalone Sensor Wearable Computer Radio Processor Processor Radio Sensor suite C Preprocessor Data Transport Network Power Distribution & Control Gateway Radio Human Interface Device Sensor suite A Public Internet Centralized Servers Future Internal Sensor Radio Processor Actuation Sensor suite B Hardware Platform Design
Processor Additional Peripherals e.g. camera Emstar Interface User Application Space OS Services interface Standard Network Access Process Manager Platform Energy Manage- ment Scheduler Network Interface Adaptive Signal Processing Automated Embedded Reasoning … Mass Storage Interface Emstar Interface Platform Energy Management Interface pulse temp gait SpO2 Preprocessor User Interface Sensor Interface Energy Manager Storage Wearable Sensors Power Distribution & Control pulse temp gait SpO2 Battery expandable/configurable programs infrastructure programs Software Design
Overview • Background • Objective diagnostics motivation • Approach • Hierarchical diagnostics • System Design • Requirements • Sensor systems • Hardware and software architecture • Implementation • Experimental results • On-node diagnostics • Next steps
Microcontroller From Motion Sensor TCP/PPP over Serial Link To Stargate From Oximeter Preprocessor • 7MHz 16-bit low power (<100mW) • 8 A/D inputs • 27Hz sampling rate for 3 channels • RTOS • C code implementation • Supports standard network interface • Example • Zworld LP3500
Processor • Intel X-Scale PXA255 • 400MHz 32-bit low power (<3W) • Multi-purpose Linux programming environment • Hosts 802.11 PCMCIA wireless card • 1GB Compact Flash card • Hosts cross compiled NETICA Bayesian Network Library • Example • Stargate
Wearable System Implementation • Wearable package • Separate battery pack • Carried in belt pack • Will be duplicated to permit monitoring of multiple individuals • Separate standalone instruments also implemented • Blood pressure cuff • Weight scale
Oximeter Data Motion
Oximeter Data Motion Rest
Oximeter Data Motion Rest Motion
Microcontroller Stargate From Server Collect additional standalone sensor data Update CPT Sample Detection Normal t Motion Data Gait Analysis Algorithm Fuse data Bayesian Network Algorithm f Boot Stargate Threshold Checking Blood Oxygen Detect Alarm f Transfer data via TCP sockets t To Server Threshold Feedback from server Secure Socket Layer Protocol (SSL) pulse rate Networking algorithm and upload data to server Alert user of alarm condition Prototype Data Acquisition
Bayesian Network Diagnostic Approach • Exploit extensive Medical Informatics experience in Bayesian automated reasoning for diagnostics • Exploit UCLA SamIam tool of Professor Darwiche • A Bayesian Network consists of two parts • A directed acyclic graph (DAG) representing influence among a set of variables • A set of conditional probability tables (CPT) that quantify these influences • The DAG model • may be constructed by domain experts • Conditional probability tables • Can be trained with historical data or assigned by domain experts
Hierarchical diagnostic pathway Disease Blood Pressure Blood Test Body Temperature Local instrument Oxygen Level Clinic Visit Pulse Rate Wearable sensors
Hierarchical diagnostic pathway P(Yes) – 70% P(No) – 30% Disease Blood Pressure Blood Test Body Temperature Local instrument Oxygen Level Clinic Visit Pulse Rate High Low High Wearable sensors
Hierarchical diagnostic pathway P(Yes) – 85% P(No) – 15% Disease Blood Pressure High Blood Test Body Temperature Local instrument Oxygen Level Clinic Visit Pulse Rate High Low High Wearable sensors
Hierarchical diagnostic pathway P(Yes) – 95% P(No) – 5% Disease Positive Blood Pressure High Blood Test Body Temperature Local instrument Oxygen Level Clinic Visit Pulse Rate High Low High Wearable sensors
Bayesian Network Gait Diagnostics Example • Sensors: • 3 accelerometer axes • (Guidance from prior work on gait analysis) • Data acquisition • Time series • Processing • Frequency domain analysis • Identify dominant frequency component • Quantize components into both frequency and amplitude states • Bayesian Network Construction • Acquire data record “case files” for walking, running, limping, and standing (specific behaviors recommended by medical informatics collaborators) • Create corresponding CPTs • Inference Proceeds
Motion Freq X Amp X Freq Y Amp Y Freq Z Amp Z Bayesian Network for Gait Diagnostics
New Evidence Bayesian Network DAG Motion State Probabilities Motion Prediction Historical Data CPT Bayesian Network DiagnosticsData Flow
Laboratory Testbed Sensors • Next steps for system training • Testbed sensors developed to enable test and verification of gait diagnostics • Resistive flex sensors • Measure knee angle • Validate methods for • CENS Summer Undergraduate program • Emstar interface
Data from Knee Flex Sensors Walking Limping
Next Steps • System design and implementation for end-to-end MEDIC system • Collaboration with Medical Informatics Team • Exploit Emstar capability • Experimental verification of MEDIC concept • Expert input on disease signatures with • Multiple wearable sensors • Local sensor and instrument data • Patient queries • Demonstrate ability of hierarchical diagnostics to detect disease conditions
Next Steps (cont) • Assist with development of server-side systems • Mark Verghese of Medical Informatics Group • Comprehensive patient records, family history, population data • Autonomous coordination and run-time management of remote MEDIC nodes • Algorithm library management • Server-side inference
Acknowledgement • UCLA Medical Informatics • Dr. Hooshang Kangarloo • Dr. Alex Bui • Dr. Ricky Tiara • SamIam team • Emstar team • Undergraduate students in EE190D • CENS Summer Undergraduate interns • Steve Liu for his work in Embedded Bayes Engine
References • Y. Shahar, C. Combi, “Timing is everything. Time-oriented clinical information systems”, West J Med 1998 Feb;168(2):105-13. • I. Korhonen, et al., “TERVA: Wellness Monitoring System”, Proceedings of the 20th Annual International Conference of the IEEE EMBS, Volume: 4, 29 Oct.-1 Nov. 1998, pp 1988-1991. • A. Lymberis, “Smart Wearable Systems for Personalized Health Management: Current R&D and Future Challenges”, Proceedings of the 25th Annual International Conference of the IEEE EMBS, Sept. 2003, pp 3716-3719. • A. Herzog, L. Lind, “Network Solutions for Home Health Care Applications”, Technology and Health Care, Vol. 11, Issue 2. Pages: 77-87. 2003. • I. Korhonen, J. Parkka, M. V. Gils, “Health Monitoring in the Home of the Future”, IEEE Engineering in Medicine and Biology Magazine, Volume: 22, Issue: 3, May-June 2003, pp 66-73. • C. Glaros, D.I. Fotiadis, A. Likas, A. Stafylopatis, “A Wearable Intelligent System for Monitoring Health Condition and Rehabilitation of Running Athletes,” 4th International IEEE EMBS Special Topic Conference on Information Technology Applications in Biomedicine, 24-26 April 2003 Pages:276-9.
References • R. Hoyt, T.S. Coster, “Combat Medical Informatics: Present and Future”, Proceedings of the AMIA Annual Symposium, 9-13 Nov. 2002 Pages: 335-9. • E. Jovanov, A. O'Donnell Lords, D. Raskovic, P.G. Cox, R. Adhami, F. Andrasik, “Stress Monitoring Using a Distributed Wireless Intelligent Sensor System”, IEEE Engineering in Medicine and Biology Magazine, Volume: 22, Issue: 3, May-June 2003 Pages: 49 – 55. • S. Park. S. Jayaraman, “Enhancing the quality of life through wearable technology”, IEEE Engineering in Medicine and Biology Magazine, Volume: 22, Issue: 3, May-June 2003 Pages:41-8 • N. Kern, B. Schiele, A. Schmidt, “Multi–sensor activity context detection for wearable computing”, Proceedings of European Symposium on Ambient Intelligence (EUSAI), Nov. 2003 Pages 220–232 • I.P.I. Pappas, T. Keller, S. Mangold, M.R. Popovic, V. Dietz, M. Morari, “A reliable gyroscope-based gait-phase detection sensor embedded in a shoe insole”,IEEE Sensors Journal, Volume: 4, Issue: 2, April 2004 Pages:268 – 274
References • S.J. Morris, J.A. Paradiso, “Shoe-integrated sensor system for wireless gait analysis and real-time feedback”, Proceedings of the Second Joint EMBS/BMES Conference, Volume: 3, 23-26 Oct. 2002 Pages:2468-9 • R. DeVaul, M. Sung, J. Gips, A. Pentland, “MIThril 2003: Applications and Architecture”, Proceedings of Seventh IEEE International Symposium on Wearable Computers, 21-23 Oct. 2003 Pages:4-11 • J.K. Pollard, S. Rohman, M.E. Fry, “A Web-based Mobile Medical Monitoring System”, IEEE International Workshop on Intelligent Data Acquisition and Advanced Computing Systems: Technology and Applications, 1-4 July 2001 Pages:32-5 • J.K. Pollard, M.E. Fry, S. Rohman et al., “Wireless and Web-based Medical Monitoring in the Home”, Med. Inform Internet Medicine 2002; Volume: 27, Issue: 3, Pages 219–27 • B. Woodward, R.S.H. Istepanian, C.I. Richards,“Design of a Telemedicine System Using a Mobile Telephone”, IEEE Transactions on Information Technology in Biomedicine, Volume: 5, Issue: 1, March 2001 Pages:13-5