Abstract:Myocardial infarction (MI) is a critical cardiovascular disease that requires a timely and accurate diagnosis to prevent severe outcomes. Clinically, MI is located according to the diagnostic criteria of the 12-lead electrocardiogram (ECG). Increasingly, various models have been proposed for MI location but most of them ignored the relevant medical domain knowledge. Furthermore, most models focus more on intra-lead features while neglecting inter-lead features. To address these issues, we propose a knowledge-guided multi-branch attention network. Based on leads grouping, clinical diagnostic knowledge is incorporated into the grouping strategy. Meanwhile, corresponding multi-branch networks are designed to accurately extract intra and inter lead features through centralized 2D residual convolution block and multi-scale 1D convolution block. Additionally, a branch attention fusion method is proposed to amplify the weight of important branches and effectively fuse features extracted by each branch. The proposed method is evaluated on the publicly available multi-label dataset PTBXL. Compared with state-of-the-art methods, it achieves significant improvements in multiple metrics, demonstrating superior performance.

KUMA-MI: A 12-Lead Knowledge-Guided Multi-branch Attention Networks for Myocardial Infarction Localization