MoE-Enhanced-TTT: Advancing Medical Image Segmentation
Chen, Yuming Wu, Qi Xie, Yutong
Chen, Yuming
Wu, Qi
Xie, Yutong
Author
Supervisor
Department
Computer Vision
Embargo End Date
Type
Conference proceeding
Date
License
Language
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
Automatic medical image segmentation has become essential in clinical applications, providing valuable support for diagnosis, treatment planning, and patient monitoring. Recent research has improved segmentation accuracy by combining convolutional approaches with sequence modeling techniques such as Transformer. However, Transformerbased methods often face significant computational challenges, especially for processing high-resolution 3D medical images. In this paper, we propose an enhanced Test-Time Training (TTT) layer, called TTT-MoE, and integrate it within the advanced nnU-Net framework. Our TTT-MoE layer harnesses a Mixture of Experts (MoE) approach to enhance the internal self-supervised learning modules originally developed for the TTT layer. This methodology enables more effective extraction of clinically relevant features, improving segmentation performance across diverse anatomical structures and imaging modalities. The code is available at github.
Citation
Y. Chen, Q. Wu, Y. Xie, "MoE-Enhanced-TTT: Advancing Medical Image Segmentation," 2025, pp. 1-8.
Source
Proceedings of the Digital Image Computing: Technqiues and Applications (DICTA)
Conference
2025 International Conference on Digital Image Computing: Techniques and Applications (DICTA)
Keywords
46 Information and Computing Sciences, 4611 Machine Learning
Subjects
Source
2025 International Conference on Digital Image Computing: Techniques and Applications (DICTA)
Publisher
IEEE