Related Experiment Video
Updated: Aug 5, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
SCAGC-UNet: Graph Convolutional Network with Spatial and Channel Attention for Medical Image Segmentation
Xiaolong Hu1, Xueyan Liu1, Junji Jiang1
1School of Mathematical and Systems Science, Liaocheng University, Liaocheng 252000, China.
Journal of Imaging
|July 27, 2026
Summary
SCAGC-UNet, a novel region-aware graph convolutional network, effectively addresses medical image segmentation challenges by integrating local details and global context. This approach demonstrates superior performance across diverse datasets, enhancing clinical diagnostic potential.
Area of Science:
- Medical image analysis
- Computer vision
- Artificial intelligence in healthcare
Background:
- Current medical image segmentation methods, including Convolutional Neural Networks (CNNs) and Transformers, face limitations in balancing local feature extraction and global context modeling.
- CNNs are restricted by local receptive fields, while Transformers can dilute semantic information when processing global context.
Purpose of the Study:
- To introduce SCAGC-UNet, a region-aware graph convolutional network designed to overcome the limitations of existing medical image segmentation techniques.
- To effectively bridge local detail extraction and global dependency modeling through structured region-level reasoning.
Main Methods:
- The SCAGC-UNet architecture employs a dual-layer residual encoder for hierarchical feature extraction.
- A Spatial-Channel Graph Convolution (SC-GCN) module at the bottleneck captures inter-region spatial topology and intra-region channel semantics.
- Context-Corrected Modules and Backward-Aided Modules are utilized in the decoder for enhanced feature refinement and reduced semantic gaps.
Main Results:
- SCAGC-UNet achieved state-of-the-art performance on three public benchmarks: Kvasir-SEG (Dice 92.28%, MIOU 92.41%), BUSI (IOU 78.10%, MIOU 87.68%), and COVID-19 CT (DSC 82.51%).
- The model consistently outperformed strong CNN-based (e.g., CCBANet) and Transformer-based (e.g., TransUNet) baselines across all datasets.
- Demonstrated robust segmentation capabilities, particularly for irregular lesion morphologies in COVID-19 CT scans.
Conclusions:
- SCAGC-UNet offers a robust and consistent solution for medical image segmentation across various modalities and challenges.
- The proposed region-aware graph convolutional approach effectively integrates local and global information, outperforming existing methods.
- The model's strong performance suggests significant potential for clinical relevance in medical image analysis and diagnosis.