Related Experiment Video
Updated: Jul 16, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
A Hybrid U-Shaped Deep Learning Network for Intracerebral Hemorrhage Segmentation in CT Scans
Ming Deng1, Jiazuo Yao1, Qingxiang Wu2
1School of Computer Science and Engineering, Guilin University of Technology, Guilin 541006, China.
Sensors (Basel, Switzerland)
|July 15, 2026
Summary
TransAMGNet, a novel deep learning model, significantly improves intracerebral hemorrhage (ICH) segmentation on CT scans. This AI-powered tool enhances lesion boundary detection for better stroke assessment and computer-aided diagnosis.
Area of Science:
- Medical Imaging and Artificial Intelligence
- Neurosurgery and Stroke Management
- Computer Vision and Deep Learning
Background:
- Computed tomography (CT) scans are crucial for rapid stroke assessment, requiring accurate segmentation of intracerebral hemorrhage (ICH).
- Existing deep learning models face challenges in segmenting complex ICH lesions due to blurred boundaries, irregular shapes, and scale variations.
- Reliable computer-aided diagnosis for ICH necessitates effective intelligent analysis of CT images.
Purpose of the Study:
- To propose TransAMGNet, a hybrid U-shaped network with Transformer integration for improved ICH CT image segmentation.
- To enhance the representation of complex ICH lesion morphology and improve segmentation accuracy.
- To overcome limitations of current deep learning methods in handling intricate lesion characteristics.
Main Methods:
- Developed TransAMGNet, a U-Net backbone integrated with a Transformer encoder for global context modeling.
- Incorporated an Adaptive Dual-branch Channel Attention Module (ADCAM) for enhanced feature sensitivity.
- Introduced a Multi-scale Feature Enhancement Module (MFEM) in skip connections and a Gate-enhanced Dynamic Upsampling Module (GDUM) for improved decoding.
Main Results:
- TransAMGNet achieved superior performance across multiple metrics: Dice (90.47%), Recall (87.83%), IoU (81.26%), Precision (91.13%), and HD95 (32.94).
- Comparative experiments demonstrated TransAMGNet's outperformance against competing methods.
- Ablation studies confirmed the significant contribution of each proposed module to the overall performance.
Conclusions:
- TransAMGNet effectively addresses the challenges of segmenting complex intracerebral hemorrhage lesions in CT images.
- The proposed network architecture and modules enhance the accuracy and reliability of AI-driven ICH segmentation.
- This advancement holds potential for improving clinical decision-making and computer-aided diagnosis in stroke management.