Related Experiment Video
Updated: Aug 15, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.9K
A deep network embedded with rough fuzzy discretization for OCT fundus image segmentation.
Qiong Chen1,2, Lirong Zeng3, Cong Lin4,5
1College of Electronic and Information Engineering, Guangdong Ocean University, Haida Road, Zhanjiang, 524000, Guangdong, China. 13907534385@163.com.
Scientific Reports
|January 7, 2023
Summary
This study introduces a novel deep network for Optical Coherence Tomography (OCT) fundus image segmentation, significantly reducing noise and improving accuracy. The Rough Fuzzy Discretization Network (RFDDN) outperforms existing methods in key segmentation metrics.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Deep learning algorithms for medical image segmentation face performance bottlenecks due to noise and redundant information.
- Accurate segmentation of Optical Coherence Tomography (OCT) fundus images is crucial for diagnosing retinal diseases.
Purpose of the Study:
- To propose a novel deep network, Rough Fuzzy Discretization Network (RFDDN), for enhanced OCT fundus image segmentation.
- To address noise and redundant information issues in deep learning-based segmentation.
Main Methods:
- Developed RFDDN integrating rough fuzzy discretization for feature preprocessing.
- Utilized fuzzy c-means clustering and a genetic algorithm for optimal feature discretization.
- Incorporated a deep supervised attention mechanism for multi-scale information extraction.
Main Results:
- RFDDN demonstrated superior performance across all evaluation indicators compared to U-Net, ReLayNet, CE-Net, MultiResUNet, and ISCLNet.
- RFDDN achieved higher Dice Similarity Coefficient (DSC), sensitivity, and specificity than ISCLNet.
- RFDDN showed lower Hausdorff Distance 95 (HD95) and Average Surface Distance (ASD) than ISCLNet.
Conclusions:
- The proposed RFDDN effectively eliminates noise and redundant information in OCT fundus images.
- RFDDN significantly improves the accuracy of OCT fundus image segmentation.
- The method balances interpretability and computational efficiency, offering a robust solution for retinal image analysis.
