Related Experiment Video
Updated: Sep 8, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.9K
Automatic segmentation of multitype retinal fluid from optical coherence tomography images using semisupervised deep
Feng Li1, WenZhe Pan2, Wenjie Xiang2
1School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai, China lifenggold@163.com.
The British Journal of Ophthalmology
|June 13, 2022
Summary
A novel deep learning model, Ref-Net, accurately segments subretinal fluid (SRF) and intraretinal fluid (IRF) in optical coherence tomography (OCT) images. This AI tool achieves expert-level performance with minimal labeled data, aiding in ocular disease management.
Area of Science:
- Ophthalmology
- Medical Imaging
- Artificial Intelligence
Background:
- Automated segmentation of retinal fluid in OCT images is crucial for diagnosing and managing various ocular diseases.
- Existing methods often require extensive labeled data, limiting their clinical applicability.
- Subretinal fluid (SRF) and intraretinal fluid (IRF) are key indicators of retinal pathology.
Purpose of the Study:
- To develop and validate a deep learning model for automated segmentation of multitype retinal fluid using OCT images.
- To assess the performance and generalizability of the proposed model.
- To investigate the impact of different components on model performance through ablation studies.
Main Methods:
- Development of a novel semisupervised retinal fluid segmentation deep network (Ref-Net).
- Training and validation on a retrospective dataset of 2814 anonymized OCT images from 141 patients.
- Testing generalizability on an unseen Kermany dataset.
- Quantitative and qualitative performance analysis using Dice similarity coefficient (Dice), sensitivity (Sen), specificity (Spe), and mean absolute error (MAE).
Main Results:
- Ref-Net achieved high performance metrics: SRF (Dice: 81.2%, Sen: 87.3%, Spe: 98.8%, MAE: 1.1%) and IRF (Dice: 78.0%, Sen: 83.6%, Spe: 99.3%, MAE: 0.5%).
- The model demonstrated expert-level performance with as few as 80 labeled OCT images and outperformed human experts with 160 labeled images.
- Satisfactory generalization capability was confirmed on an unseen dataset.
Conclusions:
- The semisupervised Ref-Net model effectively segments multitype retinal fluid using minimal labeled OCT images.
- This AI tool shows significant potential to assist clinicians in the diagnosis and management of ocular diseases.
- The Ref-Net model offers a promising solution for automated retinal fluid segmentation, improving efficiency and accuracy in clinical practice.
