Hierarchical Transformer for Electrocardiogram Diagnosis

Xiaoya Tang, JAKE AARON BERGQUIST, Benjamin A. Steinberg, Tolga Tasdizen

Keywords: Multi-scale Transformer, ECG Classification, Depth-wise Convolution

TL;DR: A hierarchical Transformer for ECG analysis that combines depth-wise convolutions, multi-scale feature aggregation, and an attention-gated module to learn inter-lead associations.

Abstract: We propose a hierarchical Transformer for ECG analysis that combines depth-wise convo- lutions, multi-scale feature aggregation via a CLS token, and an attention-gated module to preserve inter-lead relationships and enhance interpretability. The model is lightweight, flexible, and eliminates the need for complex attention or downsampling strategies. Code is available at :https://github.com/xiaoyatang/3stageFormer.git.

Submission Number: 35