40 likes | 144 Vues
This paper discusses a groundbreaking long-term physiological monitoring solution utilizing a wireless ECG system. Developed by researchers from Arizona State University and the University of Washington, it integrates low-energy consumption and efficient data storage techniques to ensure high fidelity ECG signal transmission. The model includes periodic signal updates and maintains robustness against wireless errors, making it applicable for continuous patient monitoring. The findings indicate significant improvements in battery life and data handling, enabling effective arrhythmia detection with minimal bandwidth usage.
E N D
GeM-REM Energy-efficient Long Term Physiological Monitoring AyanBanerjee, SandeepGupta Arizona State University SidharthNabar,RadhaPoovendran University of Washington, Seattle Clinical Collaborator: Dr. Barbara Cochrane, UW School of Nursing
Long Term Monitoring Solution Sensed ECG Model Feature updates ECG Leads Output ECG Match? Compare Align Sensor Platform Raw ECG samples Model Wireless Link Physician Base Station Raw signal updates 1. Low Energy 2. Small Batteries 3. Low Memory 4. Simple processing High Fidelity Data Quick Review • Storage Efficient • Data Annotation 1. Low Bandwidth Usage 2. Robust to Wireless Errors Sensor Module Base Station Module Keep alive Wireless Link
Example Operation Periodic Signal Update Low Battery Lifetime (3 hrs)[1] High Storage (7.2 GB per month) Bandwidth Usage 82 kbps Scan ______ samples per day High Fidelity Data, transmitted at 82 kbps 3 lead ECG 1.2 0.4 0 -0.4 0 200 400 600 800 1000 1200 1400 Arrhythmia probability of occurrence = 0.0097 [2] (raw signal update) Baseline and expected variations (feature updates) Baseline and expected variations (feature updates) High Fidelity Data at 82 kbps Diagnostically Equivalent Data at 1.166 kbps Diagnostically Equivalent Data at 1.166 kbps Prolonged Battery Life (126 hrs) Low Storage (175 MB per month) Low Bandwidth Usage 1.95 kbps Annotated ECG GeM-REM Reduction in data transmission Diagnostic Feature Accuracy 42:1 93% [1] Munshiet al. "Wireless ECG plaster for body sensor network," Medical Devices and Biosensors, 2008. ISSS-MDBS 2008 [2]Martinelli et al. “Probability of occurrence of life-threatening ventricular arrhythmias in Chagas' disease versus non-Chagas' disease.” PubMed
Implementation Details Generative Model: Model learned using Matlab curve fitting tool Learning Accuracy = 2.13% Devices: Shimmer ECG Google donated Nexus Computation Energy Overhead = 0.8 mW Code Size = 5KB in RAM and 22 KB in ROM Application Size = 123 KB Demo Video: http://www.youtube.com/watch?v=NGBq-oyPhGI&feature=channel_video_title