Convolutional neural network (CNN)-enabled electrocardiogram (ECG) analysis: a comparison between standard...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Artificial intelligence (AI) shows potential for early cardiac condition detection via standard 12-lead electrocardiograms (ECGs).
• The efficacy of AI in identifying cardiac abnormalities from single-lead ECGs requires systematic investigation.

Purpose of the Study:

• To assess a convolutional neural network's (CNN) performance in identifying ECG abnormalities using a single-lead (D1) setup versus a standard 12-lead setup.

Main Methods:

• A lightweight CNN was designed to detect 20 cardiac abnormalities using the PTB-XL dataset.
• The CNN accommodated various lead inputs, comparing standard 12-lead, single-lead D1, and D1 with an additional lead.

Main Results:

• The single-lead D1 CNN demonstrated satisfactory performance, with an average AUC difference of -8.7% compared to the 12-lead setup.
• For specific conditions, single-lead D1 achieved comparable diagnostic AUC to the 12-lead ECG.
• Adding a second lead to D1 reduced the AUC gap to -2.8% versus the 12-lead setup.

Conclusions:

• A lightweight CNN can predict cardiac abnormalities from single-lead D1 ECGs, similar to 12-lead ECGs.
• Findings support the use of single-lead ECGs from wearable devices for large-scale cardiac abnormality screening.