Convolution-Free Waveform Transformers for Multi-Lead ECG Classification

TL;DR

This paper introduces a convolution-free waveform transformer model for multi-lead ECG classification, demonstrating competitive performance in detecting cardiac abnormalities across various ECG lead subsets in a large challenge dataset.

Contribution

The paper presents a novel waveform transformer architecture for ECG classification and evaluates its performance across multiple lead subsets, achieving competitive challenge rankings.

Findings

Achieved an average challenge metric of 0.47 across all ECG-lead subsets.

Ranked 11th out of 39 teams in the PhysioNet/CinC challenge.

Performed consistently across different ECG-lead configurations.

Abstract

We present our entry to the 2021 PhysioNet/CinC challenge - a waveform transformer model to detect cardiac abnormalities from ECG recordings. We compare the performance of the waveform transformer model on different ECG-lead subsets using approximately 88,000 ECG recordings from six datasets. In the official rankings, team prna ranked between 9 and 15 on 12, 6, 4, 3 and 2-lead sets respectively. Our waveform transformer model achieved an average challenge metric of 0.47 on the held-out test set across all ECG-lead subsets. Our combined performance across all leads placed us at rank 11 out of 39 officially ranking teams.