What does this research mean for the field?

The RhythmX self-supervised contrastive learning framework achieves precision and recall between 0.99 and 0.999 in beat classification and macro-F1 scores of 0.91 to 0.94 on external ECG datasets. Novelty: ClaimNovelty.METHODOLOGICAL. Consensus alignment: ConsensusAlignment.NEUTRAL.

What question did this study set out to answer?

The aim is to enhance ECG arrhythmia classification through a self-supervised learning framework that retains noisy segments.

March 3, 2026Open Access

RhythmX™: An Interpretable Self-Supervised Contrastive Learning Framework for Heartbeat Classification

Key Result

The RhythmX™ framework achieved precision and recall between 0.99–0.999 in beat classification and macro-F1 scores of 0.91–0.94 on external ECG datasets.

Key Points

The aim is to enhance ECG arrhythmia classification through a self-supervised learning framework that retains noisy segments.
Developed a self-supervised ECG representation learning framework combining contrastive pretraining and ensemble classification.
Applied a signal-to-noise ratio criterion during pretraining to stabilize optimization.
Implemented a hybrid ensemble model with convolutional encoders and tree-based classifiers.
Executed patient-level partitioning with stratified 10-fold cross-validation for model evaluation.
Performed external validation on independent ECG datasets such as PTB-XL and Chapman–Shaoxing.
Achieved high precision and recall values between 0.99 and 0.999 across various ECG beat categories.
Obtained macro-F1 scores ranging from 0.91 to 0.94 on external validation datasets.
Demonstrated strong performance compared to standalone convolutional and Random Forest classifiers.

Structured PICO

Does the RhythmX self-supervised contrastive learning framework improve automated ECG heartbeat classification compared to standalone convolutional and Random Forest classifiers?

Population

ECG datasets including PTB-XL, Chapman-Shaoxing, and INCART 12-lead datasets representing heterogeneous patient populations and recording configurations

Intervention

RhythmX, a self-supervised ECG representation learning framework combining contrastive pretraining with ensemble-based supervised classification

Comparator

Standalone convolutional and handcrafted feature-based Random Forest classifiers

Outcome

Beat-level classification performance (precision, recall, macro-F1 score)

The RhythmX self-supervised learning framework demonstrates high performance and generalizability for automated ECG heartbeat classification across diverse patient cohorts.

Main Result

Absolute Event Rate: 0% vs 0%

Abstract

Automated electrocardiogram (ECG) arrhythmia classification remains challenging due to signal noise, inter-patient variability, and limited annotated data, which constrain the generalization of supervised learning approaches. This study presents a self-supervised ECG representation learning framework that combines contrastive pretraining with ensemble-based supervised classification. A signal-to-noise ratio criterion is applied during self-supervised pretraining to stabilize contrastive optimization, while all extracted ECG beats, including noisy segments, are retained during downstream evaluation. The learned representations are classified using a hybrid ensemble composed of convolutional encoders and tree-based models. Model evaluation follows strict patient-level partitioning with stratified 10-fold cross-validation and bootstrap-based uncertainty estimation on a held-out test set. Under this evaluation protocol, the framework achieved high beat-level performance on curated datasets (internal and external). Class-wise performance shows precision and recall values between 0.99 and 0.999 across normal, supraventricular, ventricular, fusion, and paced beat categories. External validation is conducted on independent ECG cohorts, including PTB-XL, Chapman–Shaoxing, and INCART 12-lead datasets. On these datasets, the hybrid model attains macro-F1 scores ranging from 0.91 to 0.94, compared with standalone convolutional and handcrafted feature-based Random Forest classifiers evaluated under identical conditions. These results characterize the behavior of the proposed representation learning framework across heterogeneous patient populations and recording configurations.