Bidirectional Fusion Guided by Cardiac Patterns for Semi-Supervised ECG Segmentation
A new framework called CardioMix has been introduced for ECG segmentation, addressing the challenge of limited annotated data. This framework utilizes a bidirectional CutMix strategy guided by cardiac patterns to enhance both labeled and unlabeled data. Extensive experiments show that CardioMix outperforms existing methods in ECG delineation across various datasets.
- ▪Accurate ECG segmentation is crucial for cardiovascular diagnostics but is hindered by a lack of annotated data.
- ▪CardioMix enriches labeled data with variations from unlabeled data while maintaining physiological relevance.
- ▪The framework is compatible with various semi-supervised segmentation algorithms and consistently outperforms existing strategies.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Machine Learning arXiv:2605.15722 (cs) [Submitted on 15 May 2026] Title:Bidirectional Fusion Guided by Cardiac Patterns for Semi-Supervised ECG Segmentation Authors:Jeonghwa Lim, Minje Park, Sunghoon Joo View a PDF of the paper titled Bidirectional Fusion Guided by Cardiac Patterns for Semi-Supervised ECG Segmentation, by Jeonghwa Lim and 2 other authors View PDF HTML (experimental) Abstract:Accurate delineation of electrocardiogram (ECG), the segmentation of meaningful waveform features, is crucial for cardiovascular diagnostics. However, the scarcity of annotated data poses a significant challenge for training deep learning models.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
