Related Experiment Video
Updated: Aug 11, 2025
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
Automatic Segmentation of Hyperreflective Foci in OCT Images Based on Lightweight DBR Network
Jin Wei1,2, Suqin Yu1, Yuchen Du1,3
1Department of Ophthalmology, Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine, National Clinical Research Center for Eye Diseases, Shanghai Key Laboratory of Ocular Fundus Diseases, Shanghai Engineering Center for Visual Science and Photomedicine, Shanghai Engineering Center for Precise Diagnosis and Treatment of Eye Diseases, Shanghai, 200080, China.
We developed a fast, lightweight neural network for segmenting hyperreflective foci (HF) in optical coherence tomography (OCT) images. This automated method aids in diagnosing fundus diseases like diabetic macular edema (DME), improving prognosis and clinical practice.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Hyperreflective foci (HF) in optical coherence tomography (OCT) images indicate inflammatory responses in fundus diseases such as diabetic macular edema (DME), retina vein occlusion (RVO), and central serous chorioretinopathy (CSC).
- Accurate HF segmentation is crucial for disease prognosis, but traditional methods are slow and computationally intensive.
- Developing automated, efficient HF segmentation is vital for clinical applications.
Purpose of the Study:
- To propose and evaluate a lightweight neural network for rapid and accurate automatic segmentation of hyperreflective foci (HF) in OCT images.
- To compare the proposed method's performance against existing techniques in terms of speed and accuracy.
- To assess the generalizability of the proposed network across different fundus diseases.
Main Methods:
- A two-stage framework was employed, starting with an NLM filter and patch-based split for image preprocessing.
- A lightweight DBR neural network was utilized for the automatic segmentation of HF.
- The framework was trained and tested on 3000 OCT images from 300 patients with DME, RVO, and CSC.
Main Results:
- The proposed lightweight DBR network achieved a segmentation speed of 57 ms per OCT image, significantly faster than traditional methods.
- The DBR network demonstrated high performance with area under the curve dice similarity coefficients (DSC) of 83.65% (DME), 76.43% (RVO), and 82.20% (CSC).
- The method achieved at least a 5% higher DSC compared to previous approaches and showed easier segmentation for HF in DME.
Conclusions:
- The proposed lightweight DBR network offers a fast and accurate solution for automatic HF segmentation in OCT images.
- This method is universally applicable to various fundus diseases, showing potential for widespread clinical adoption.
- The efficiency and accuracy of the DBR network can significantly aid in the prognosis and management of inflammatory fundus diseases.
