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

The reconstruction of a standard 12-lead ECG from a reduced number of leads is a key challenge for wearable ECG devices, which typically record only 1 to 3 leads. This study explores the feasibility of reconstructing the full 12-lead ECG using deep learning models trained on a reduced set of three leads, extracted from a body surface potential map (BSPM) with 35 electrodes. Universal models were employed across all subjects to enhance computational efficiency and ensure generalizability, eliminating the need for subject-specific training. A total of 30 models were trained by combining multiple three-lead input configurations with two different architectures: convolutional-only and convolutional-temporal models. The two best-performing models achieved median R values of 0.98 and 0.97 across all leads. The findings highlight the potential of deep learning models for accurate and efficient 12-lead ECG reconstruction, with future research focusing on extending the model to pathological populations.Clinical relevance- This study highlights the potential of deep learning to reconstruct 12-lead ECGs from three leads, enhancing the diagnostic capabilities of wearable ECG devices for continuous cardiac monitoring.