Related Experiment Video
Updated: Jun 14, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
CDSegNet: a multi-scale convolution and attention mechanism U-shaped network for Crohn's disease lesion segmentation
Peipei Wang1, Yu Liu2, Yuanjun Wang3
1School of Health Science and Engineering, University of Shanghai for Science and Technology, 516 Jungong Road, Yangpu District, Shanghai, 200093, Shanghai, China.
Radiological Physics and Technology
|June 12, 2026
Summary
This study introduces CDSegNet, a novel U-Net based model for segmenting Crohn's disease (CD) lesions in computed tomography enterography (CTE) images. CDSegNet significantly improves lesion segmentation accuracy, aiding in diagnosis and personalized treatment.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Gastroenterology
Background:
- Accurate segmentation of Crohn's disease (CD) lesions from computed tomography enterography (CTE) is vital for diagnosis and treatment planning.
- Existing studies on automatic CD lesion segmentation are limited, highlighting a need for improved computational methods.
Purpose of the Study:
- To propose a novel deep learning model, CDSegNet, for enhanced automatic segmentation of CD lesions in CTE images.
- To improve the accuracy and robustness of CD lesion segmentation, particularly for varying lesion sizes.
Main Methods:
- Development of CDSegNet, a U-Net based model incorporating residual dilated and standard convolution feature extraction (RDCFE) and residual attention feature extraction (RAFE) modules.
- Integration of a multi-scale convolution module to aggregate features from diverse receptive fields.
- Quantitative performance assessment using Intersection over Union (IoU), Recall, Dice Similarity Coefficient (DSC), and Hausdorff distance (HD).
Main Results:
- CDSegNet achieved high performance metrics: Recall (0.867), IoU (0.820), HD (23.28), and DSC (0.895).
- The model demonstrated significant improvements over the baseline U-Net, with IoU, HD, and DSC increasing by 11.6%, 3.21, and 10.4%, respectively.
- CDSegNet showed particular strengths in segmenting small and medium lesions and maintained stable performance across all test images.
Conclusions:
- CDSegNet presents a competitive and effective approach for automated Crohn's disease lesion segmentation in CTE.
- The model's ability to refine features and handle varying lesion sizes offers potential benefits for clinical diagnosis and treatment personalization.
- Further clinical validation is recommended before widespread adoption in practice.
