Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

SemiSegECG: A Multi-Dataset Benchmark for Semi-Supervised Semantic Segmentation in ECG Delineation

Created by
  • Haebom

Author

Minje Park, Jeonghwa Lim, Taehyung Yu, Sunghoon Joo

Outline

This paper focuses on electrocardiogram (ECG) segmentation, which segments meaningful features from electrocardiogram (ECG) waveforms. Because advances using deep learning have been limited by the lack of publicly available annotated datasets, semi-supervised learning leveraging rich unlabeled ECG data presents a promising solution. In this study, we present SemiSegECG, the first systematic benchmark for semi-supervised semantic segmentation (SemiSeg) in ECG segmentation. We curate and integrate multiple public datasets, including previously untapped sources, to enable robust and diverse evaluations. We employ five representative SemiSeg algorithms from computer vision, implement them on two different architectures—Convolutional Neural Networks (CNNs) and Transformers—and evaluate them in both within-domain and cross-domain settings. We also propose ECG-specific training configurations and augmentation strategies, and introduce a standardized evaluation framework. Our results demonstrate that Transformers outperform CNNs in semi-supervised ECG segmentation. SemiSegECG is expected to serve as a foundation for advancing semi-supervised ECG segmentation methods and stimulating further research in this field.

Takeaways, Limitations

•
Takeaways:
◦
We present the first systematic ECG segmentation benchmark for semi-supervised semantic segmentation, called SemiSegECG.
◦
Integrating various public datasets enables robust and diverse evaluations.
◦
We demonstrate that the transformer architecture outperforms convolutional neural networks in semi-supervised ECG segmentation.
◦
Contributes to performance improvement by suggesting ECG-specific training configurations and augmentation strategies.
◦
It provides a foundation for contributing to the research and development of semi-supervised ECG segmentation methods.
•
Limitations:
◦
Generalization performance may vary depending on the characteristics and scale of the dataset used.
◦
Performance comparisons of other algorithms other than the five presented SemiSeg algorithms may be necessary.
◦
Additional performance validation in actual clinical environments is required.
◦
Additional analysis of performance differences across domains may be required.
👍
Made with Slashpage