Hierarchical Transformer for Electrocardiogram Diagnosis

TL;DR

This paper introduces a hierarchical Transformer model for ECG diagnosis that effectively captures inter-lead relationships, improves interpretability, and remains lightweight without complex attention mechanisms.

Contribution

The proposed model uniquely combines depth-wise convolutions, multi-scale feature aggregation, and attention gating to enhance ECG analysis and interpretability.

Findings

Achieves accurate ECG classification with fewer parameters.

Effectively models inter-lead relationships.

Provides improved interpretability of ECG features.

Abstract

We propose a hierarchical Transformer for ECG analysis that combines depth-wise convolutions, multi-scale feature aggregation via a CLS token, and an attention-gated module to learn inter-lead relationships and enhance interpretability. The model is lightweight, flexible, and eliminates the need for complex attention or downsampling strategies.