Related Experiment Video
Updated: Jul 8, 2026
06:48
Automated Segmentation of Cortical Grey Matter from T1-Weighted MRI Images
Published on: January 7, 2019
A lightweight network for segmenting tree-like structures in medical images.
Irfan Dwiki Bhaswara1, Sonit Singh1, Arcot Sowmya1
1School of Computer Science and Engineering, University of New South Wales, Sydney, NSW, Australia.
Summary
A new deep learning model, CoorMa, effectively segments complex tree-like structures in medical images. It overcomes limitations of prior methods by using Coordinate Attention and Mamba, achieving superior accuracy with fewer parameters.
Area of Science:
- Medical Image Analysis
- Deep Learning
- Computational Anatomy
Background:
- Accurate segmentation of tree-like structures (e.g., vasculature, airways) is vital for medical applications.
- Existing deep learning methods, particularly CNNs and Transformers, face challenges with long-range dependencies and computational complexity in segmenting these structures.
- Mamba offers a promising alternative with linear complexity for capturing long-range dependencies.
Purpose of the Study:
- To introduce CoorMa, a novel, lightweight U-Net-based model for segmenting tree-like structures in medical images.
- To enhance segmentation accuracy by integrating Coordinate Attention and Mamba Aggregation Modules.
- To improve robustness by incorporating a skeleton-based artefact removal algorithm.
Main Methods:
- Developed CoorMa, a U-Net-shaped model incorporating Coordinate Attention for directional feature enhancement.
- Integrated Mamba Aggregation Modules to effectively fuse multi-scale features and capture long-range dependencies.
- Implemented a skeleton-based artefact removal technique to refine segmentation results.
Main Results:
- CoorMa demonstrated superior performance across four 2D datasets (DCA1, CHUAC, DRIVE, CHASE_DB1) and one 3D dataset (Binary Airway Segmentation).
- The model achieved state-of-the-art evaluation metrics, outperforming existing methods.
- CoorMa achieved these results with a significantly reduced number of model parameters, indicating high efficiency.
Conclusions:
- CoorMa presents an effective and efficient solution for segmenting challenging tree-like structures in medical imaging.
- The integration of Coordinate Attention and Mamba modules significantly enhances the model's ability to handle complex anatomical structures.
- The proposed method offers a promising advancement for applications requiring precise medical image segmentation.