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.

Comparative Analysis of CNN and Transformer Architectures with Heart Cycle Normalization for Automated Phonocardiogram Classification

Created by
  • Haebom

Author

Martin Sondermann, Pinar Bisgin, Niklas Tschorn, Anja Burmann, Christoph M. Friedrich

Outline

This paper systematically compares and analyzes the performance of four models (two CNNs and two BEATs) for cardiac noise detection. It is evaluated on the PhysioNet2022 dataset using fixed-length and cardiac period normalization methods, and a new cardiac period normalization method tailored to individual heart rhythms is proposed. The CNN model achieves an AUROC of 79.5% in the fixed-length method and 75.4% in the cardiac period normalization method. The BEATs model achieves an AUROC of 65.7% in the fixed-length method and 70.1% in the cardiac period normalization method. The results suggest that the regularization strategy has a significant impact on the model performance, and the balance between accuracy and computational efficiency is important in clinical settings. Although the CNN model shows higher accuracy, the BEATs model has the advantage of development efficiency in terms of fast learning and evaluation speed.

Takeaways, Limitations

•
Takeaways:
◦
We reveal that cardiac period normalization strategy has a significant impact on the performance of PCG classification models.
◦
A specific CNN architecture has been shown to perform well in cardiac noise detection.
◦
The zero-shot BEATs model can provide advantages in terms of development efficiency.
◦
An automated PCG classification system has the potential to contribute to improved cardiac diagnosis and patient management.
•
Limitations:
◦
This study was limited to a specific dataset (PhysioNet2022) and may have limited generalizability.
◦
Additional analysis of model performance for various heart diseases and heart murmur types is needed.
◦
The accuracy of the BEATs model is lower than that of the CNN model, requiring further research on practical clinical applications.
👍
Made with Slashpage