Reducing leads, enhancing wearable practicality: A comparative study of 3-lead vs. 12-lead ECG classification

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Device Technology

Background:

• Traditional 12-lead electrocardiograms (ECGs) are essential for diagnosing cardiac conditions but can be costly and complex.
• Wearable ECG devices are increasing in popularity, driving interest in simplified ECG analysis.
• Deep learning models show promise for automated ECG interpretation.

Purpose of the Study:

• To evaluate the feasibility of using reduced-lead ECGs (specifically 3-lead) for automated detection of heart anomalies using deep learning.
• To adapt and optimize a state-of-the-art 12-lead deep learning model for 3-lead ECG data.
• To assess the diagnostic accuracy of a 3-lead ECG model compared to a 12-lead model for classifying cardiac pathologies.

Main Methods:

• Adapted a 12-lead deep learning model (Ribeiro et al.) for 3-lead configurations by modifying the input layer.
• Trained the model architecture from scratch on the public PTB-XL database.
• Optimized the 3-lead model using transfer learning and a One-vs-All classification approach for a five-class setup (normal, myocardial infarction, ST/T change, conduction disturbance, hypertrophy).

Main Results:

• The initial 3-lead model, despite a 75% reduction in input data, showed only a minor 3% performance drop compared to the 12-lead model.
• The novel optimized 3-lead model achieved a micro-averaged F1-score of 77%.
• The 12-lead model achieved a micro-averaged F1-score of 78%, indicating near-equivalent performance of the optimized 3-lead model.

Conclusions:

• Simplified reduced-lead ECG classification models can achieve diagnostic accuracy comparable to traditional 12-lead ECGs.
• Deep learning models adapted for 3-lead ECGs offer a cost-effective and accessible approach for cardiac diagnostics.
• This advancement has the potential to democratize early cardiac diagnostics, particularly in resource-limited settings.