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.

Motion-enhanced Cardiac Anatomy Segmentation via an Insertable Temporal Attention Module

Created by
  • Haebom

Author

Md. Kamrul Hasan, Guang Yang, Choon Hwai Yap

Outline

This paper proposes the Temporal Attention Module (TAM) to improve the accuracy of cardiac anatomy segmentation. Existing cardiac segmentation methods that utilize cardiac motion information suffer from limitations such as high computational costs and poor robustness. TAM is designed as a lightweight, multi-head cross-temporal attention module that can be easily applied to various networks (CNN-based, Transformer-based, or hybrid) used in various cardiac imaging systems (2D and 3D ultrasound and MRI). Experiments on various datasets demonstrate that TAM outperforms existing methods while maintaining computational efficiency. TAM provides a robust and generalized solution that scales from 2D to 3D.

Takeaways, Limitations

•
Takeaways:
◦
We propose a lightweight, plug-and-play temporal attention module (TAM) and demonstrate its easy applicability to various cardiac image segmentation networks.
◦
Achieves improved accuracy and computational efficiency compared to existing methods.
◦
It has scalability applicable to both 2D and 3D images.
◦
Provides a generalized solution applicable to various network structures.
•
Limitations:
◦
Further studies are needed to determine whether the proposed TAM performs optimally across all cardiac imaging datasets and network architectures.
◦
Further validation of applicability and utility in real clinical settings is needed.
◦
Further performance evaluations for various types of heart disease are needed.
👍
Made with Slashpage