Related Experiment Video
Updated: Jun 30, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.7K
A novel Dual-Branch Asymmetric Encoder-Decoder Segmentation Network for accurate colonic crypt segmentation
Jingjun Zhou1, Hong Xiong1, Qian Liu2
1School of Biomedical Engineering, Hainan University, Haikou, 570228, China.
Computers in Biology and Medicine
|March 24, 2024
Summary
A new AI model, DAUNet, accurately segments colonic crypts (CC) in images, crucial for diagnosing colorectal cancer (CRC). This advanced segmentation aids physicians in precise lesion identification and analysis, improving CRC diagnostic accuracy.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Oncology
- Gastroenterology
Background:
- Colorectal cancer (CRC) is a major cause of cancer mortality.
- Accurate segmentation of colonic crypts (CC) is vital for CRC diagnosis and treatment strategies.
- Existing automatic segmentation methods face challenges due to blurred boundaries and diverse morphology of CC.
Purpose of the Study:
- To develop a novel and efficient deep learning model for accurate segmentation of CC in confocal laser endomicroscopy (CLE) images.
- To address the challenges of blurred boundaries and morphological diversity in CC segmentation.
- To improve the precision of lesion localization and region analysis for CRC diagnosis.
Main Methods:
- Proposed the Dual-Branch Asymmetric Encoder-Decoder Network (DAUNet) for CC segmentation.
- Introduced a dual-branch feature extraction module (DFEM) with Focus operations and dense depth-wise separable convolution (DDSC) for multiscale feature extraction.
- Incorporated a feature fusion guided module (FFGM) with cross-group spatial and channel attention for adaptive feature fusion and a local multi-layer perceptron (LMLP) for feature refinement.
Main Results:
- DAUNet achieved high Intersection over Union (IoU) scores of 81.54% and 84.83% on two independent datasets.
- The model demonstrated effectiveness comparable to state-of-the-art methods in CC segmentation.
- Experimental results validate the model's capability in handling diverse CC morphologies and blurred boundaries.
Conclusions:
- The proposed DAUNet model offers a robust and effective solution for automatic CC segmentation in CLE images.
- This technology has significant potential to assist clinicians in precise lesion identification and analysis, thereby enhancing CRC diagnostic accuracy.
- Further development and application of DAUNet could lead to improved patient outcomes in colorectal cancer management.
