Effective 12-Lead ECG Reconstruction from Minimal Lead Sets Using Deep Learning for Advanced Wearable Systems.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Medicine
• Cardiology

Background:

• Wearable electrocardiogram (ECG) devices face challenges in reconstructing a standard 12-lead ECG due to limited lead recordings (1-3 leads).
• Body Surface Potential Mapping (BSPM) provides high-resolution data, but its application in wearable devices is limited.

Purpose of the Study:

• To investigate the feasibility of reconstructing a 12-lead ECG from a reduced set of three leads using deep learning.
• To develop universal deep learning models for efficient and generalizable ECG reconstruction across subjects.

Main Methods:

• Trained 30 deep learning models using three-lead configurations extracted from 35-electrode BSPM data.
• Employed convolutional-only and convolutional-temporal architectures for model development.
• Utilized universal models without subject-specific training for enhanced efficiency and generalizability.

Main Results:

• The two best-performing models achieved high median R values of 0.98 and 0.97 across all leads.
• Demonstrated accurate and efficient reconstruction of 12-lead ECGs from limited leads.
• Validated the generalizability of universal deep learning models.

Conclusions:

• Deep learning models show significant potential for accurate 12-lead ECG reconstruction from three leads.
• This approach can enhance the diagnostic capabilities of wearable ECG devices for continuous cardiac monitoring.
• Future research should focus on extending these models to pathological populations.