Detecting COVID-19 from digitized ECG printouts using 1D convolutional neural networks

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• The COVID-19 pandemic highlighted global healthcare system vulnerabilities, necessitating rapid and cost-effective diagnostic tools.
• Cardiac injury is a reported complication of COVID-19, suggesting electrocardiograms (ECG) as potential biomarkers.
• Existing diagnostic methods for COVID-19 can be time-consuming and resource-intensive.

Purpose of the Study:

• To develop an automated system for COVID-19 detection using electrocardiogram (ECG) signals.
• To propose and validate a novel method for extracting usable ECG data from paper records.
• To assess the efficacy of a deep learning model in diagnosing COVID-19 from digitized ECG signals.

Main Methods:

• A novel technique was developed to digitize ECG signals from traditional paper records.
• The quality of digitized signals was evaluated by comparing R-peak and RR-interval measurements against original paper ECGs.
• A one-dimensional convolutional neural network (1D-CNN), specifically SEResNet18, was trained on the digitized ECG signals for classification.

Main Results:

• The proposed digitization method accurately captured original ECG signals, achieving a mean absolute error of 28.11 ms.
• The SEResNet18 model demonstrated high accuracy in classifying COVID-19: 98.42% (vs. Normal) and 98.50% (vs. other classes).
• The system also performed effectively in multi-classification tasks, indicating robust diagnostic capability.

Conclusions:

• Digitized ECG signals can reliably represent original cardiac electrical activity.
• A deep learning system trained on digitized ECGs shows significant potential as a tool for automated COVID-19 diagnosis.
• This approach offers a promising avenue for rapid, cost-effective, and accessible COVID-19 screening and diagnosis.