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.

BenchECG and xECG: a benchmark and baseline for ECG foundation models

Created by
  • Haebom

Author

Riccardo Lunelli, Angus Nicolson, Samuel Martin Proll , Sebastian Johannes Reinstadler, Axel Bauer, Clemens Dlaska

Outline

This paper presents BenchECG, a standardized benchmark for electrocardiogram (ECG) data, and xECG, an ECG base model based on xLSTM trained with SimDINOv2 self-supervised learning. BenchECG addresses the problem of task selection bias and dataset inconsistency seen in previous studies by incorporating various publicly available ECG datasets and tasks, enabling fair comparison. xECG outperforms existing state-of-the-art models on BenchECG, and is the only publicly available model that demonstrates robust performance across all datasets and tasks. Therefore, BenchECG accelerates the advancement of ECG representation learning, and xECG sets a new standard for future ECG base models.

Takeaways, Limitations

•
Takeaways:
◦
Presentation of BenchECG, a standardized benchmark for fair comparison of ECG basic models.
◦
We present a new baseline model, xECG, that performs well across all datasets and tasks.
◦
Accelerating the Advancement of ECG Representation Learning Research
◦
Presenting a new performance standard for future ECG basic model research.
•
Limitations:
◦
Additional clarification is needed regarding the types and scope of datasets and tasks included in BenchECG.
◦
Lack of detailed description of the specific structure and learning process of the xECG model.
◦
Lack of comparative analysis with other self-supervised learning methods
◦
Validation of xECG model performance in real clinical environments is needed.
👍
Made with Slashpage