Research
Atrial Fibrillation Detection on the Embedded Edge: Energy-Efficient Inference on a Low-Power Microcontroller
Akbari, Y., Lei, N., Patel, N., Peng, Y., & Faust, O. (2025). Atrial Fibrillation Detection on the Embedded Edge: Energy-Efficient Inference on a Low-Power Microcontroller. Sensors, 25(21), 6601.
The problem, in plain English
Atrial fibrillation is a common heart arrhythmia that's easy to miss because it's often intermittent — you need continuous monitoring to catch it, which means the monitoring device has to be small, cheap, and battery-friendly enough to wear all the time. This work puts a deep-learning model directly on a low-cost microcontroller so detection happens on-device, in real time, without sending data to the cloud.
Technical approach
- Model: Bi-LSTM network, trained and validated against the MIT-BIH Arrhythmia Database
- Target hardware: STM32L475 (Arm Cortex-M4, low-power MCU class)
- Full on-device inference — no cloud round-trip, no external compute
Results
Accuracy98.46%
Inference time143ms
Power draw24.7mW
Validated against the MIT-BIH Arrhythmia Database.
Cite this work
@article{akbari2025afdetection,
author = {Akbari, Yash and Lei, Ning and Patel, Nirali and Peng, Yan and Faust, Oliver},
title = {Atrial Fibrillation Detection on the Embedded Edge: Energy-Efficient Inference on a Low-Power Microcontroller},
journal = {Sensors},
volume = {25},
number = {21},
pages = {6601},
year = {2025},
doi = {10.3390/s25216601}
}