Detection of Disease Features on Retinal OCT Scans Using RETFound

Area of Science:

• Ophthalmology
• Artificial Intelligence
• Medical Imaging

Background:

• Age-related macular degeneration (AMD) and other eye diseases cause irreversible vision loss.
• Early detection is crucial for managing retinal conditions.
• Optical coherence tomography (OCT) is vital for visualizing retinal pathology, but manual interpretation is time-consuming and inconsistent.

Purpose of the Study:

• To automate the classification of key disease signatures in OCT images using a foundation model.
• To evaluate the performance of the RETFound model against ResNet-50 for retinal disease diagnosis.

Main Methods:

• Leveraged RETFound, a foundation model pretrained on 1.6 million unlabeled OCT images.
• Finetuned RETFound and compared its performance with ResNet-50 on a dataset of 1770 labeled OCT B-scans.
• Classified features including subretinal fluid (SRF), intraretinal fluid (IRF), drusen, and pigment epithelial detachment (PED).

Main Results:

• RETFound models achieved accuracy ranging from 0.75 to 0.77 and AUC-ROC values from 0.75 to 0.80.
• Performance was comparable to ResNet-50 in terms of specificity and sensitivity.
• Both single-task and multitask modes were explored for RETFound.

Conclusions:

• The RETFound model demonstrates potential as a tool for automated retinal disease diagnosis from OCT scans.
• It offers comparable performance to established models like ResNet-50.
• RETFound may enhance diagnostic accuracy and interpretability, supporting clinicians in OCT image analysis.