Related Experiment Video
Updated: Feb 28, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.6K
DCANet: Disentanglement and Category-Aware Aggregation for Medical Image Segmentation
Xiaoqing Li1, Hua Huo1, Chen Zhang1
1Information Engineering College, Henan University of Science and Technology, Luoyang 471000, China.
Sensors (Basel, Switzerland)
|February 27, 2026
Summary
DCANet, a novel framework for medical image segmentation, improves accuracy by integrating local and global features and enhancing category awareness. This approach effectively addresses challenges posed by ambiguous boundaries and complex anatomy.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Background:
- Accurate medical image segmentation is crucial for clinical applications.
- Challenges include ambiguous boundaries and complex anatomical structures.
Purpose of the Study:
- To introduce DCANet (Disentangled and Category-Aware Network) for improved medical image segmentation.
- To effectively integrate local and global features and enhance category-aware interactions.
Main Methods:
- Feature fusion using Feature Coupling Unit (FCU) to combine local and global information.
- Decoupled Feature Module (DFM) to separate foreground and background features.
- Category-Aware Integration Aggregator (CAIA) for multi-level feature fusion and boundary refinement.
Main Results:
- DCANet achieved superior performance on four public datasets (Synapse, ACDC, GlaS, MoNuSeg).
- Reported Dice scores: 84.80% (Synapse), 94.07% (ACDC), 94.60% (GlaS), and 79.85% (MoNuSeg).
Conclusions:
- DCANet demonstrates effectiveness and generalizability in segmenting complex anatomical structures.
- The framework successfully resolves boundary ambiguities in diverse medical image segmentation tasks.
