Atrial Fibrillation Detection on the Embedded Edge: Energy-Efficient Inference on a Low-Power Microcontroller.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Embedded Systems

Background:

• Atrial Fibrillation (AF) is a prevalent cardiac arrhythmia often undiagnosed, leading to severe health risks like stroke and heart failure.
• Current AF detection methods may require complex equipment or cloud-based processing, posing challenges for continuous, remote monitoring.
• There is a need for efficient, low-power solutions for real-time AF detection directly on edge devices.

Purpose of the Study:

• To develop and evaluate a novel Embedded Edge system for real-time Atrial Fibrillation (AF) detection.
• To demonstrate the feasibility of performing AF classification on a low-power Microcontroller Unit (MCU) using optimized AI models.
• To enable energy-efficient, privacy-preserving, and scalable cardiac monitoring outside traditional clinical settings.

Main Methods:

• Extraction of Heart Rate Variability (HRV) features from RR-Interval (RRI) data.
• Implementation of a compact Long Short-Term Memory (LSTM) model optimized for embedded deployment on an MCU.
• Real-time classification of AF events directly on the edge device without relying on full ECG or cloud analytics.

Main Results:

• Achieved an overall classification accuracy of 98.46% for AF detection.
• Inference completed in 143 ± 0 ms with minimal power consumption of 3532 ± 6 μJ per inference on the target MCU.
• Demonstrated feasibility of local inference enabling strategic wireless communication for alerts, enhancing privacy and battery life.

Conclusions:

• Clinically meaningful AF monitoring is achievable on constrained edge devices.
• The developed system offers a practical solution for energy-efficient, privacy-preserving, and scalable AF screening.
• This work advances personalized and decentralized cardiac care through practical AI-driven edge diagnostics.