A wearable ultra-low power real-time ECG analysis system based on spiking neural network

To address the challenge of balancing high power consumption and real-time performance in long-term monitoring of wearable Electrocardiogram (ECG) monitoring devices, this paper proposes an edge intelligence system integrated with a Spiking Neural Network (SNN). By embedding a lightweight SNN model in the terminal, the system enables localized real-time analysis of ECG signals and arrhythmia detection, avoiding the drawbacks of continuous wireless transmission of raw data. The core innovations of the system include: 1) an SNN classification model based on Leaky Integrate-and-Fire (LIF) neurons and the surrogate gradient training method; 2) a synergistically optimized ultra-low power hardware architecture incorporating a dedicated SNN coprocessor. Experiments conducted on the Massachusetts Institute of Technology-Beth Israel Hospital Arrhythmia Database (MIT-BIH) Arrhythmia Database demonstrate that the system maintains a high classification accuracy of 98.2%, while reducing the inference power consumption for a single heartbeat classification by approximately 90% compared to traditional Convolutional Neural Network (CNN) models. The overall system power consumption is decreased by 79.6%, and the theoretical battery life is significantly extended from 2 months to over 10 months. This research provides an effective technical pathway for the development of next-generation long-lasting wearable medical devices.