3-Lead to 12-Lead ECG Reconstruction: A Novel AI-based Spatio-Temporal Method

TL;DR

This paper introduces a novel AI-based LSTM-UNet model that effectively reconstructs 12-lead ECG signals from reduced lead data, improving accuracy over traditional methods and aiding portable cardiac monitoring.

Contribution

The paper presents a combined LSTM-UNet model that captures both temporal and spatial dependencies for ECG reconstruction, outperforming existing linear and LSTM-based methods.

Findings

Achieved 94.37% R^2 on PhysioNet PTBDB database.

Outperformed LSTM and linear transformation methods.

Demonstrated improved ECG reconstruction accuracy.

Abstract

Diagnosis of cardiovascular diseases usually relies on the widely used standard 12-Lead (S12) ECG system. However, such a system could be bulky, too resource-intensive, and too specialized for personalized home-based monitoring. In contrast, clinicians are generally not trained on the alternative proposal, i.e., the reduced lead (RL) system. This necessitates mapping RL to S12. In this context, to improve upon traditional linear transformation (LT) techniques, artificial intelligence (AI) approaches like long short-term memory (LSTM) networks capturing non-linear temporal dependencies, have been suggested. However, LSTM does not adequately interpolate spatially (in 3D). To fill this gap, we propose a combined LSTM-UNet model that also handles spatial aspects of the problem, and demonstrate performance improvement. Evaluated on PhysioNet PTBDB database, our LSTM-UNet achieved a mean R^2 value of 94.37%, surpassing LSTM by 0.79% and LT by 2.73%. Similarly, for PhysioNet INCARTDB database, LSTM-UNet achieved a mean R^2 value of 93.91%, outperforming LSTM by 1.78% and LT by 12.17%.