FPGA-based Low Precision Neural Network for Arrhythmia detection

Contents

• Introduction
• Requirements
• Usage
• FAQ and Issues
• Acknowledgement
• Citation

Introduction

This project is aimed at deploying accurate, unbiased neural networks, which can be used in hospitals to detect arrhythmia at early stages, with processing on low-power, cheap and edge hardware to preserve privacy of patients.

Neural Net Architecture

Requirements

• An ubuntu PC or VM(8GB RAM minimum) with sudo access (Tested on Ubuntu 20.04 Focal Fossa)
• A PYNQ supported board connected to the internet via ethernet (Tested on PYNQ-Z2)
• SD Card (16GB minimum)
• Ethernet cable
• Micro USB cable

Note: For Installing and setting up environment refer to SETUP.md

Usage

cd /path/to/finn
./run-docker.sh notebook
cd notebooks
git clone git@github.com:Adityasrinivas24/ecg-classification-pynq.git

Install required packages outside docker environment

pip3 install -r requirements.txt

Pretrained Models

Model	Specification	ONNX
EcgNet	W2A2	ECGNet W2A2
EcgNet	W4A4	EcgNet W4A4
HeartNet	W4A4	Heartnet W4A4

Notebooks

Outside the Docker Environment
1-ecgnet-train-test trains and validates a quantized neural network to classify ecg images

In the Docker Environment
2-ecgnet-onnx-export exports the model to finn compatible onnx format
3-ecgnet-finn-compile has all the steps required to compile design using finn

On PYNQ
4-ecgnet-pynq-deploy contains details on pynq deployment and results

FAQ and Issues

Resolve the NewConnectionError raised by pip during library installations in PYNQ refer this discussion

Other FAQs and Issues can be found here

Acknowledgement

@software{brevitas,
  author       = {Alessandro Pappalardo},
  title        = {Xilinx/brevitas},
  year         = {2022},
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.3333552},
  url          = {https://doi.org/10.5281/zenodo.3333552}
}

@inproceedings{finn,
author = {Umuroglu, Yaman and Fraser, Nicholas J. and Gambardella, Giulio and Blott, Michaela and Leong, Philip and Jahre, Magnus and Vissers, Kees},
title = {FINN: A Framework for Fast, Scalable Binarized Neural Network Inference},
booktitle = {Proceedings of the 2017 ACM/SIGDA International Symposium on Field-Programmable Gate Arrays},
series = {FPGA '17},
year = {2017},
pages = {65--74},
publisher = {ACM}
}