Deep learning model for ECG reconstruction reveals the information content of ECG leads

Objective: This study aimed to evaluate the information content of individual electrocardiogram (ECG) leads and their inter-lead correlations using a deep learning approach. Specifically, we investigated the capability of a neural network to reconstruct missing ECG leads from reduced-lead configurations, thereby revealing each lead's unique and shared informational value. Methods: We developed a U-net convolutional neural network to reconstruct missing leads in 12-lead ECG recordings. The model was trained using the PTB-XL dataset and tested on the PTB dataset. We trained the model with varying combinations of input leads, including single limb leads, combinations of two limb leads and configurations, including one or two precordial leads. We evaluated the model's performance using mean squared error (MSE) between the reconstructed and actual signals. Results: The models demonstrated varying reconstruction accuracy depending on the input lead configuration. Precordial leads V1 and V6 showed the highest reconstruction fidelity from limb leads alone, while V3 consistently exhibited the lowest, indicating its unique informational content. Conclusions: The proposed method effectively quantifies the informational value of ECG leads. This has significant implications for optimizing lead selection in diagnostic scenarios, particularly in settings where complete 12-lead ECGs are impractical. In addition, the study provides insights into the physiological underpinnings of ECG signals and their propagation. The findings pave the way for advances in telemedicine, portable ECG devices and personalized cardiac diagnostics by reducing redundancy and improving signal interpretation.