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

2025-06-24

AI Summary

ECG analysis suffers from “simplicity bias”—a tendency of supervised models to over-rely on high-frequency, coarse-grained patterns while neglecting clinically critical fine-grained abnormalities. This work provides the first empirical evidence of this phenomenon and proposes a time-frequency dual-domain multi-granularity self-supervised learning framework. It enhances dynamic feature modeling via time-frequency-aware filtering and jointly learns coarse- and fine-grained representations through masked reconstruction and contrastive learning-driven prototype reconstruction—without requiring manual annotations, thereby mitigating overfitting to explicit patterns. Evaluated on six public ECG datasets across three downstream diagnostic tasks (e.g., arrhythmia classification, ischemia detection), the method achieves state-of-the-art performance, notably improving accuracy in detecting subtle pathological patterns. Additionally, we release a large-scale, multicenter ECG dataset to advance the field.

Technology Category

• Machine Learning: Unsupervised & Self-Supervised Learning
• Computer Vision: Multi-modal Vision
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Web Mining and Content Analysis: Large pretrained models with web data

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. Code and dataset will be released publicly.

Problem

• Demonstrates Simplicity Bias in ECG models overlooking critical cues
• Proposes self-supervised learning to mitigate diagnostic performance issues
• Introduces temporal-frequency filters for multi-grained ECG representation

Innovation

• Temporal-Frequency aware Filters for ECG dynamics
• Multi-Grained Prototype Reconstruction for dual domains
• Self-supervised learning to mitigate Simplicity Bias