A Deformable Lead-Attention Fusion Network for Multi-Label ECG Classification Integrating Clinical Metadata

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Abstract Cardiovascular diseases remain a major global health burden, making accurate elec-trocardiogram (ECG) analysis essential for timely diagnosis. While deep learning–based methods have made significant progress in automating ECG classification, they are often limited by the variability of ECG signals, the frequent presence of multiple cardiac abnor-malities, and the inadequate integration of diverse clinical data. This study proposes DLM-Net: a robust deep learning framework for multi-label ECG classification, designed to address these challenges. DLM-Net in-corporates several key innovations. First, the Lead-Specific Deformable Path applies de-formable convolutions independently to each lead. This enables adaptive modelling of non-rigid and complex waveform variations. It is also enhanced by a High-Frequency Module, which captures fine-grained tem-poral details. Secondly, a spatial path based on a ResNet-like backbone and feature pyr-amid network (FPN) fuses multi-scale repre-sentations in order to effectively capture both global and local ECG patterns. In parallel, the Lead Path uses a Lead Attention Trans-former to explicitly model inter-lead de-pendencies. Finally, ECG features from the spatial and lead paths are integrated with clinical tabular data through a multimodal fusion module, in which compact pa-tient-level tabular embeddings provide com-plementary contextual information for ECG representation. Extensive experiments on three public multi-label ECG datasets PTB-XL, CPSC2018 and Chapman demonstrate that DLM-Net achieves compet-itive and stable performance across datasets, with label-wise accuracies of 89.11%, 96.05% and 97.82%, respectively. Overall, DLM-Net provides an effective and flexible framework for multi-label ECG classification, offering improved accuracy and robustness across diverse datasets and demonstrating the effec-tiveness of the proposed architecture in han-dling complex multi-label diagnostic tasks in retrospective settings.