Real-Time Myocardial Infarction Detection Approaches with a Microcontroller-Based Edge-AI Device.

Area of Science:

• Biomedical Engineering
• Cardiology
• Computer Science

Background:

• Myocardial Infarction (MI), or heart attack, requires continuous monitoring due to its recurring and often asymptomatic nature.
• Wearable devices are crucial for real-time health surveillance, but often rely on cloud connectivity for data processing.
• Edge computing offers a decentralized approach for on-device data analysis, reducing latency and enhancing privacy.

Purpose of the Study:

• To develop and evaluate a single-microcontroller system for automatic Myocardial Infarction (MI) detection using edge computing.
• To compare the performance of Machine Learning (ML) and Deep Learning (DL) algorithms optimized for low-resource hardware.
• To assess the feasibility of real-time MI detection on a 32-bit microcontroller with an ARM Cortex-M4 core.

Main Methods:

• Implementation of two MI detection algorithms: one based on ML with feature extraction and a simpler Neural Network (NN), and another based on DL using Spectrogram Analysis and a Convolutional Neural Network (CNN).
• Optimization of both ML and DL algorithms for deployment on low-resource, single-microcontroller hardware (ARM Cortex-M4).
• Feasibility assessment including accuracy, inference time, and memory usage analysis for both approaches.

Main Results:

• The ML approach achieved 89.40% accuracy, while the DL approach reached 94.76% accuracy, both utilizing the same low-power hardware.
• The DL method, while having a longer inference time and higher memory usage, demonstrated superior detection accuracy.
• The developed system performs all processing at the edge, enabling real-time MI detection without cloud or remote server reliance.

Conclusions:

• A single-microcontroller edge computing system can effectively perform real-time Myocardial Infarction detection.
• Deep learning models, despite higher resource demands, offer superior accuracy for MI detection on resource-constrained devices.
• The energy-efficient prototype enables continuous, private, and immediate MI monitoring via wearable technology.