Deep Learning for Cardiologist-level Myocardial Infarction Detection in Electrocardiograms

EasyChair Preprint 1543

Abstract

Heart disease is the leading cause of death worldwide. Amongst patients with cardiovascular diseases, myocardial infarction is the main cause of death. Thus, detection of myocardial infarction in a timely manner is a serious challenge with a significant potential for impact. Here, we study the impact of multiple channels of observation to correctly classify heart conditions, finding that lead I and lead II are critical to obtain correct classifications using data from the Physikalisch-Technische Bundesanstalt (PTB) database. Based on these findings, we develop a convolutional neural network to detect myocardial infarction using lead I and lead II electrocardiogram (ECG) signals. Our approach differs from others in the community in that it does not require any kind of manual feature extraction or pre-processing of any kind. Rather, the raw ECG signal is fed into the neural network. When evaluated, the model achieves a 99.15% accuracy, reaching cardiologist-level performance level for myocardial infarction detection. Preliminary experiments indicate that coupling this neural network model with a denoising deep learning model increases classification accuracy even further.

Keyphrases: Applications of AI, computational biology, machine learning, statistical learning

@booklet{EasyChair:1543,
  author    = {Arjun Gupta and Eliu Huerta and Issam Moussa},
  title     = {Deep Learning for Cardiologist-level Myocardial Infarction Detection in Electrocardiograms},
  howpublished = {EasyChair Preprint 1543},
  year      = {EasyChair, 2019}}