Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

BEAT-Net: Injecting Biomimetic Spatio-Temporal Priors for Interpretable ECG Classification

Runze Ma, Caizhi Liao

TL;DR

BEAT-Net addresses data inefficiency and opacity in ECG diagnosis by tokenizing signals into heartbeat units and applying a biomimetic, multi-encoder architecture. By explicitly separating local morphology, lead-specific spatial context, and rhythm, the method achieves competitive accuracy with a fraction of annotated data and offers interpretable attention aligned with clinical heuristics. Across PTB-XL, CPSC2018, and CSN, BEAT-Net matches dominant 1D-CNN baselines while demonstrating strong data efficiency (≈30–35% annotations) and improved interpretability, underscoring the value of biological priors in ECG analysis. Overall, this approach provides a scalable, efficient, and medically intelligible alternative to heavy pre-training for ECG classification.

Abstract

Although deep learning has advanced automated electrocardiogram (ECG) diagnosis, prevalent supervised methods typically treat recordings as undifferentiated one-dimensional (1D) signals or two-dimensional (2D) images. This formulation compels models to learn physiological structures implicitly, resulting in data inefficiency and opacity that diverge from medical reasoning. To address these limitations, we propose BEAT-Net, a Biomimetic ECG Analysis with Tokenization framework that reformulates the problem as a language modeling task. Utilizing a QRS tokenization strategy to transform continuous signals into biologically aligned heartbeat sequences, the architecture explicitly decomposes cardiac physiology through specialized encoders that extract local beat morphology while normalizing spatial lead perspectives and modeling temporal rhythm dependencies. Evaluations across three large-scale benchmarks demonstrate that BEAT-Net matches the diagnostic accuracy of dominant convolutional neural network (CNN) architectures while substantially improving robustness. The framework exhibits exceptional data efficiency, recovering fully supervised performance using only 30 to 35 percent of annotated data. Moreover, learned attention mechanisms provide inherent interpretability by spontaneously reproducing clinical heuristics, such as Lead II prioritization for rhythm analysis, without explicit supervision. These findings indicate that integrating biological priors offers a computationally efficient and interpretable alternative to data-intensive large-scale pre-training.

BEAT-Net: Injecting Biomimetic Spatio-Temporal Priors for Interpretable ECG Classification

TL;DR

BEAT-Net addresses data inefficiency and opacity in ECG diagnosis by tokenizing signals into heartbeat units and applying a biomimetic, multi-encoder architecture. By explicitly separating local morphology, lead-specific spatial context, and rhythm, the method achieves competitive accuracy with a fraction of annotated data and offers interpretable attention aligned with clinical heuristics. Across PTB-XL, CPSC2018, and CSN, BEAT-Net matches dominant 1D-CNN baselines while demonstrating strong data efficiency (≈30–35% annotations) and improved interpretability, underscoring the value of biological priors in ECG analysis. Overall, this approach provides a scalable, efficient, and medically intelligible alternative to heavy pre-training for ECG classification.

Abstract

Although deep learning has advanced automated electrocardiogram (ECG) diagnosis, prevalent supervised methods typically treat recordings as undifferentiated one-dimensional (1D) signals or two-dimensional (2D) images. This formulation compels models to learn physiological structures implicitly, resulting in data inefficiency and opacity that diverge from medical reasoning. To address these limitations, we propose BEAT-Net, a Biomimetic ECG Analysis with Tokenization framework that reformulates the problem as a language modeling task. Utilizing a QRS tokenization strategy to transform continuous signals into biologically aligned heartbeat sequences, the architecture explicitly decomposes cardiac physiology through specialized encoders that extract local beat morphology while normalizing spatial lead perspectives and modeling temporal rhythm dependencies. Evaluations across three large-scale benchmarks demonstrate that BEAT-Net matches the diagnostic accuracy of dominant convolutional neural network (CNN) architectures while substantially improving robustness. The framework exhibits exceptional data efficiency, recovering fully supervised performance using only 30 to 35 percent of annotated data. Moreover, learned attention mechanisms provide inherent interpretability by spontaneously reproducing clinical heuristics, such as Lead II prioritization for rhythm analysis, without explicit supervision. These findings indicate that integrating biological priors offers a computationally efficient and interpretable alternative to data-intensive large-scale pre-training.
Paper Structure (18 sections, 4 equations, 4 figures, 2 tables)

This paper contains 18 sections, 4 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Biological Tokenization
  4. BEAT-Net Architecture
  5. Word Encoder
  6. Spatial Encoder
  7. Temporal Encoder
  8. Sentence Encoder
  9. Experiments
  10. Datasets
  11. Baselines
  12. Implementation
  13. Results
  14. Baselines Comparison
  15. Ablation Study
  16. ...and 3 more sections

Figures (4)

  • Figure 1: Overview of the proposed framework. Stage 1 discretizes multi-lead ECG $X$ into heartbeat tokens $\mathcal{H}$. Stage 2 BEAT-Net processes tokens via a stratified pipeline.
  • Figure 2: Impact of input tokenization strategies on model performance. The bar chart illustrates the AUC improvements achieved by switching from raw waveform patching to QRS tokenization.
  • Figure 3: Data efficiency analysis on CPSC2018 and CSN. Performance curves demonstrate that BEAT-Net outperforms fully supervised baselines using only 35% of training samples.
  • Figure 4: Spatial attention patterns on PTB-XL. (a) Lead II dominance in Rhythm tasks. (b) V1, V5, and V6 localization in Form tasks. (c) Balanced lead utilization for the 'All' task. (d) Consistent distribution across classification levels.