Real-Time Cardiac Arrhythmia Classification Using TinyML on Ultra-Low-Cost Microcontrollers: A Feasibility Study for...

Area of Science:

• Embedded Systems Engineering
• Artificial Intelligence
• Biomedical Signal Processing

Background:

• Edge computing and Tiny Machine Learning (TinyML) enable AI on resource-constrained devices.
• Deploying AI on microcontrollers requires efficient models and hardware optimization.
• Cardiac arrhythmia detection is crucial for remote and low-cost healthcare solutions.

Purpose of the Study:

• To design, implement, and validate a real-time cardiac arrhythmia classification system.
• To deploy a quantized 1D-CNN model on an 8-bit Arduino UNO microcontroller.
• To demonstrate the feasibility of TinyML for low-cost cardiac monitoring.

Main Methods:

• ECG signal acquisition using AD8232, signal preprocessing, and heartbeat segmentation.
• Development and quantization of a 1D-CNN model using TensorFlow and TensorFlow Lite.
• Deployment on an Arduino UNO with real-time visualization on an OLED display.

Main Results:

• Achieved 97.6% accuracy in classifying Normal, Ventricular, and Supraventricular arrhythmias.
• Model footprint below 24 KB with an average inference time of 200 ms per heartbeat.
• Successfully enabled real-time operation on a resource-constrained microcontroller.

Conclusions:

• Lightweight deep learning models can be deployed on ultra-constrained embedded systems via TinyML.
• The system serves as a proof-of-concept for low-cost cardiac monitoring technologies.
• Highlights the trade-offs between AI performance and hardware limitations in embedded systems.