A Tiny Matched Filter-Based CNN for Inter-Patient ECG Classification and Arrhythmia Detection at the Edge

Area of Science:

• Cardiology
• Machine Learning
• Edge Computing

Background:

• Cloud-based machine learning (ML) for electrocardiogram (ECG) analysis faces limitations in availability and privacy for real-time arrhythmia detection.
• Edge inference offers a promising alternative but requires computationally efficient ML models suitable for resource-constrained devices.

Purpose of the Study:

• To develop a tiny convolutional neural network (CNN) classifier for real-time ECG monitoring at the edge.
• To address the computational demands of ML algorithms on edge devices for arrhythmia detection.

Main Methods:

• A tiny convolutional neural network (CNN) classifier was designed utilizing matched filter (MF) theory for ECG analysis.
• The model was trained and tested on the MIT-BIH dataset and validated on INCART, QT, and PTB databases.
• Performance was evaluated for generalization and in the presence of noise.

Main Results:

• The proposed classifier achieved an average accuracy of 98.18%, sensitivity of 91.90%, and F1 score of 92.17%.
• High sensitivity was observed for detecting supraventricular (85.3%) and ventricular (96.34%) ectopic beats.
• The model boasts a small size (15 KB) and rapid inference time (<1 ms), outperforming state-of-the-art ECG classifiers.

Conclusions:

• The developed tiny CNN classifier provides a computationally efficient and accurate solution for edge-based real-time ECG arrhythmia monitoring.
• Its minimal complexity and superior performance make it suitable for deployment on resource-constrained edge devices, potentially benefiting millions of cardiovascular patients.