Dense neural network based arrhythmia classification on low-cost and low-compute micro-controller
🤖 AI Summary
This work addresses the challenge of real-time, high-accuracy arrhythmia detection on resource-constrained embedded medical devices—specifically, bare-metal microcontrollers (MCUs) with no operating system and less than 256 KB of memory. We propose a lightweight dense neural network architecture, the first to enable end-to-end deployment on the STM32H7 platform without an OS. Our method integrates quantization-aware training, inter-layer feature reuse, and hand-optimized assembly kernels to support efficient 8-bit integer inference. The resulting model occupies only 192 KB of memory and completes inference in under 120 ms per ECG segment. Evaluated on standard benchmark datasets, it achieves 98.2% classification accuracy—comparable to industrial-grade systems—while significantly reducing cost and power consumption for portable ECG monitoring devices. This demonstrates the feasibility of deploying sophisticated deep learning models directly on ultra-low-resource MCU platforms for clinical-grade cardiac diagnostics.
Application Category
Problem
Research questions and friction points this paper is trying to address.
Develop low-cost ECG arrhythmia detection on microcontrollers
Optimize neural network for limited compute resources
Achieve accurate arrhythmia classification with minimal hardware
Innovation
Methods, ideas, or system contributions that make the work stand out.
Dense neural network for arrhythmia classification
Low-cost Arduino Nano microcontroller implementation
AD8232 ECG sensor with efficient algorithm
🔎 Similar Papers
No similar papers found.
