Non-invasive ECG Image Detection with Deep Learning on Edge Hardware

This project develops a deep-learning system for non-invasive ECG image classification and detection, deployed on an embedded NVIDIA Jetson microcontroller for online, real-time screening. The goal is an affordable, portable device that reads an ECG image and flags signs associated with cardiac conditions, bringing model inference to the edge rather than relying on a server.

Aims and objectives

Find a reliable deep-learning method for ECG image classification and detection: investigate suitable architectures, collect substantial open-source ECG image datasets for training and testing, and refine the selected algorithm for a software and hardware interface that is efficient in computation time and complexity.
Build a hardware framework for online ECG image detection: choose an appropriate microcontroller for the software and hardware interface, build a device that detects cardiology conditions from the trained model, and evaluate its performance for online detection.

Proposed software and hardware pipeline: CNN training and testing feeding a model deployed to an NVIDIA Jetson Nano for live ECG prediction — The proposed pipeline: a convolutional neural network is trained and tested on ECG image classes (normal, sudden death, arrhythmia, supraventricular arrhythmia), then deployed to an NVIDIA Jetson Nano for live prediction.

A convolutional neural network is trained on ECG images that are pulse-extracted and reshaped, then the trained model runs on the embedded Jetson Nano so that live data captured at the device is classified on the spot. In the published realization, the model reached up to 95% training accuracy and classified four ECG categories, including a perfect score on the sudden-death prediction set.

Research team

Dr Wahyu Caesarendra, UBD (Principal Investigator)
Dr Daphne Teck Ching Lai, UBD (Co-Principal Investigator)
Taufiq Aiman Hishamuddin, UBD (Project Member)
Asmah Husaini, UBD (Project Member)
Associate Professor Adam Glowacz, AGH University of Science and Technology, Poland (external collaborator)
Dr Mohamad Ezam Emran, RIPAS Hospital (external collaborator)

Publications

Caesarendra, W., Hishamuddin, T. A., Lai, D. T. C., Husaini, A., Nurhasanah, L., Glowacz, A., and Alfarisy, G. A. F. (2022). An Embedded System Using Convolutional Neural Network Model for Online and Real-Time ECG Signal Classification and Prediction. Diagnostics, 12(4), 795. doi:10.3390/diagnostics12040795

This open-access article is the published outcome of the project, expanding the team’s 2021 conference work on CNN-based ECG classification.

Project details

Started 1 July 2020, duration 24 months. A previous IADA research project.

Explore more on the Projects page.