Validating the Clinical Utility of CineECG 3D Reconstructions through Cross-Modal Feature Attribution

2026-04-29

AI Summary

Current deep learning models for electrocardiogram (ECG) interpretation often lack clinical interpretability, making it difficult to precisely link waveform features to underlying anatomical pathology. To address this limitation, this work proposes a cross-modal mapping approach that integrates feature attributions from standard 12-lead ECG models into the CineECG three-dimensional anatomical space. This method uniquely combines the diagnostic capability of conventional ECG analysis with the intuitive clarity of 3D visualization and introduces a cross-modal averaging mechanism to mitigate attribution instability. Experimental evaluation on a dataset of 20 expert-annotated cases demonstrates that the proposed approach significantly enhances the reliability of pathological localization, achieving a Dice coefficient of 0.56—substantially outperforming the 0.47 obtained with traditional 12-lead attribution methods.

Abstract

Deep learning models for 12-lead electrocardiogram (ECG) analysis achieve high diagnostic performance but lack the intuitive interpretability required for clinical integration. Standard feature attribution methods are limited by the inherent difficulty in mapping abstract waveform fluctuations to physical anatomical pathologies. To resolve this, we propose a cross-modal method that projects feature attributions from high-performance 12-lead ECG models onto the CineECG 3D anatomical space. Our study reveals that while models trained directly on CineECG signals suffer from reduced accuracy and incoherent attributions, the proposed mapping mechanism effectively recovers clinically relevant feature rankings. Validated against a ground-truth dataset of 20 cases annotated by domain experts, the mapped explanations yield a Dice score of 0.56, significantly outperforming the 0.47 baseline of standard 12-lead attributions. These findings indicate that cross-modal averaging mapping effectively filters attribution instability and improves the localization of pathological features, combining the diagnostic expressiveness of standard ECG with the intuitive clarity of anatomical visualization.

Problem

• ECG interpretability
• cross-modal attribution
• anatomical localization
• clinical utility
• feature attribution

Innovation

• cross-modal attribution
• CineECG
• 3D anatomical visualization
• feature attribution
• interpretable AI