Arrhythmia Detection Model

A lightweight ECG-based arrhythmia detection model built with a CNN architecture, trained in PyTorch, and optimized (via quantization and ONNX export) to run on an Arduino Nano 33 BLE — a microcontroller with only 1 MB of flash and 256 KB of RAM.

---

Overview

This project tackles the challenge of running real-time cardiac arrhythmia classification directly on an edge device. The pipeline covers everything from ECG signal preprocessing to model training, export, and embedded deployment.

The model classifies ECG signals into arrhythmia categories and is compressed to fit within the strict memory constraints of the Arduino Nano 33 BLE using model quantization.

---

Repository Structure

Arrhythmia-detection-model/
├── main.py               # Training and evaluation entry point
├── model.py              # CNN model architecture definition
├── preprocessing.py      # ECG signal preprocessing utilities
├── requirements.txt      # Python dependencies
├── best_ecg_model.pth    # Best trained PyTorch model weights
├── ecg_model.onnx        # Exported ONNX model (for TFLite/C conversion)
└── output_dir/           # Training outputs, logs, and exported artifacts

---

Model Architecture

The model is a Convolutional Neural Network (CNN) designed for 1D ECG time-series classification. It is intentionally kept compact to enable deployment on microcontrollers after quantization.

Key design decisions:

• 1D convolutions for temporal ECG feature extraction
• Minimal parameter count to fit within embedded memory budgets
• Quantization-aware design for post-training quantization

---

Pipeline

Raw ECG Signal
      ↓
preprocessing.py  →  Filtering, normalization, segmentation
      ↓
model.py          →  CNN classifier
      ↓
main.py           →  Training loop, evaluation, model export
      ↓
ecg_model.onnx    →  ONNX export
      ↓
Arduino Nano 33 BLE  →  Edge inference (via TFLite / C array)

---

Getting Started

Prerequisites

• Python 3.8+
• Arduino Nano 33 BLE (for embedded deployment)
• Arduino IDE with TensorFlow Lite Micro library

Installation

git clone https://github.com/adii11001/Arrhythmia-detection-model.git
cd Arrhythmia-detection-model
pip install -r requirements.txt

Training

python main.py

The best model weights will be saved to best_ecg_model.pth and the ONNX export to ecg_model.onnx.

---

Deployment on Arduino Nano 33 BLE

To deploy on the Arduino Nano 33 BLE:

1. Convert ecg_model.onnx to TensorFlow Lite format using ONNX → TF → TFLite toolchain.
2. Apply post-training quantization (INT8) to reduce model size.
3. Convert the .tflite model to a C byte array using xxd:

   xxd -i ecg_model.tflite > ecg_model_data.cc

4. Include the generated C array in your Arduino sketch alongside the TensorFlow Lite Micro library.
5. Flash to the Arduino Nano 33 BLE via Arduino IDE.

---

Requirements

Key Python dependencies (see requirements.txt for full list):

• torch — Model training
• onnx / onnxruntime — Model export and validation
• numpy — Signal processing
• scipy — ECG filtering utilities
• scikit-learn — Evaluation metrics