Reliable ECG Anomaly Detection on Edge Devices for Internet of Medical Things Applications

Area of Science:

• Medical technology
• Artificial Intelligence
• Embedded Systems

Background:

• Resource-constrained edge devices limit the deployment of machine learning for real-time monitoring in the Internet of Medical Things (IoMT).
• Continuous health monitoring via electrocardiogram (ECG) analysis is crucial for detecting anomalies like arrhythmias.
• Existing cloud-dependent solutions face challenges with latency, privacy, and power consumption.

Purpose of the Study:

• To introduce a novel Tiny Machine Learning (TinyML) approach for real-time ECG anomaly detection on low-power embedded systems.
• To demonstrate the feasibility of deploying optimized AI models directly on edge devices, eliminating the need for cloud connectivity.
• To achieve a balance between model accuracy and resource utilization for continuous, long-term wearable health monitoring.

Main Methods:

• Integration of TinyML with edge Artificial Intelligence (AI) on platforms like Raspberry Pi and Arduino.
• Implementation of a workflow including data preprocessing, feature extraction, and model inference executed on the edge device.
• Application of advanced optimization techniques such as model pruning and quantization to reduce memory and power consumption.

Main Results:

• Successful deployment of optimized TinyML models on edge devices for real-time ECG anomaly detection.
• Achieved an accuracy of 92.3% for ECG anomaly detection.
• Reduced power consumption to 0.024 mW, enabling energy-efficient, continuous monitoring.

Conclusions:

• TinyML integrated with edge AI offers a viable solution for real-time medical monitoring on resource-constrained devices.
• The proposed method effectively detects ECG anomalies with high accuracy and significantly low power consumption.
• This advancement paves the way for next-generation wearable health technology and continuous patient monitoring.