Related Experiment Video
Updated: Jan 1, 2026
08:22
Application of Optical Coherence Tomography to a Mouse Model of Retinopathy
Published on: January 12, 2022
5.0K
Deep learning-based automated detection of retinal diseases using optical coherence tomography images
Feng Li1, Hua Chen1, Zheng Liu1
1School of Optical-Electrical and Computer Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China.
Biomedical Optics Express
|December 20, 2019
Summary
An ensemble of ResNet50 models accurately classifies retinal diseases like diabetic macular edema in OCT images, matching ophthalmologist performance. This approach is effective for limited medical imaging data.
Area of Science:
- Ophthalmology
- Medical Imaging
- Computer-Aided Diagnosis
Background:
- Accurate retinal disease classification is crucial for computer-aided diagnosis (CAD).
- Distinguishing between choroidal neovascularization (CNV), diabetic macular edema (DME), DRUSEN, and normal retinal OCT images presents a significant challenge.
Purpose of the Study:
- To develop and evaluate an automated system for classifying four common retinal conditions using optical coherence tomography (OCT) images.
- To assess the performance of an ensemble of improved ResNet50 models for retinal disease detection.
Main Methods:
- An ensemble of four ResNet50-based classification models was developed.
- A patient-level 10-fold cross-validation was performed on a retinal OCT image dataset.
- Model predictions were qualitatively evaluated using occlusion testing to understand decision-making.
Main Results:
- The proposed ensemble achieved high classification accuracy (0.973), sensitivity (0.963), and specificity (0.985) at the B-scan level.
- Performance metrics, including AUC and kappa value, were analyzed.
- The model's performance matched or exceeded that of experienced ophthalmologists.
Conclusions:
- Multi-ResNet50 ensembling is a valuable technique for retinal disease classification, especially with limited medical image availability.
- Occlusion testing provided insights into model decision-making and misclassification patterns.
- Integrating patient medical history with OCT images may further enhance model performance.
