Memory-Augmented Hierarchical Perception for Domain-Generalized Medical Image Segmentation.
Saved in:
| Title: | Memory-Augmented Hierarchical Perception for Domain-Generalized Medical Image Segmentation. |
|---|---|
| Authors: | Mi, Yingna1 yingnami@emails.bjut.edu.cn, Shi, Meihui2 shimeihui@bjut.edu.cn, Yang, Ruiling3 yangruiling@emails.bjut.edu.cn, Zhao, Nan3 zhaonan@emails.bjut.edu.cn |
| Source: | Engineering Letters. Jun2026, Vol. 34 Issue 6, p2285-2295. 11p. |
| Subjects: | Image segmentation, Probabilistic generative models, Machine learning |
| Abstract: | Domain generalization remains a fundamental challenge in medical image segmentation due to the domain shift. In this paper, we propose a novel segmentation framework, termed Memory-augmented Hierarchical Perception (MHP), which integrates strong prior knowledge from foundation models. Specifically, we design a frozen SAM-assisted encoder to extract high-quality local features using the SAM without additional fine-tuning. To further enhance domain-invariant representation learning, we introduce a hierarchical perception module that fuses complementary features from three pathways: local semantic cues, global contextual dependencies via Vision Mamba, and anatomical boundary sensitivity through edge-aware attention. Besides, a memory-augmented imagination generator is proposed to simulate unknown-domain features by the variational generation process with a learnable codebook, enabling the model to generalize beyond the training distribution. The encoder extracts multi-level features, which are integrated by the perception module and augmented by the imagination generator. These components form the core of our MHP framework. Extensive experiments on Fundus and Prostate segmentation datasets demonstrate that our MHP outperforms existing SOTA domain generalization methods. [ABSTRACT FROM AUTHOR] |
| Copyright of Engineering Letters is the property of International Association of Engineers (IAENG) and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.) | |
| Database: | Engineering Source |
Be the first to leave a comment!
