Masked Autoencoders that Feel the Heart: Unveiling Simplicity Bias for ECG Analyses

TL;DR

This paper identifies the presence of simplicity bias in ECG analysis models, demonstrates its negative effects, and proposes a self-supervised learning approach with novel components to mitigate this bias and improve diagnostic accuracy.

Contribution

It empirically reveals simplicity bias in ECG models, and introduces a self-supervised learning method with temporal-frequency filters and multi-grained reconstruction to reduce bias and enhance performance.

Findings

Simplicity bias negatively impacts ECG diagnostic accuracy.

Self-supervised learning alleviates simplicity bias in ECG models.

Proposed method achieves state-of-the-art results across multiple ECG datasets.

Abstract

The diagnostic value of electrocardiogram (ECG) lies in its dynamic characteristics, ranging from rhythm fluctuations to subtle waveform deformations that evolve across time and frequency domains. However, supervised ECG models tend to overfit dominant and repetitive patterns, overlooking fine-grained but clinically critical cues, a phenomenon known as Simplicity Bias (SB), where models favor easily learnable signals over subtle but informative ones. In this work, we first empirically demonstrate the presence of SB in ECG analyses and its negative impact on diagnostic performance, while simultaneously discovering that self-supervised learning (SSL) can alleviate it, providing a promising direction for tackling the bias. Following the SSL paradigm, we propose a novel method comprising two key components: 1) Temporal-Frequency aware Filters to capture temporal-frequency features reflecting the dynamic characteristics of ECG signals, and 2) building on this, Multi-Grained Prototype Reconstruction for coarse and fine representation learning across dual domains, further mitigating SB. To advance SSL in ECG analyses, we curate a large-scale multi-site ECG dataset with 1.53 million recordings from over 300 clinical centers. Experiments on three downstream tasks across six ECG datasets demonstrate that our method effectively reduces SB and achieves state-of-the-art performance.