A robust deep learning classifier for screening multiple retinal diseases on optical coherence tomography.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Retinal diseases cause significant visual impairment globally.
• Early diagnosis and management are crucial to prevent blindness.
• Current AI models for ocular disease screening lack robustness and generalizability on external datasets.

Purpose of the Study:

• To develop a robust and generalizable deep learning architecture for retinal disease detection using optical coherence tomography (OCT) images.
• To address limitations of existing AI models in handling variable OCT data characteristics and native resolution processing.
• To improve the clinical applicability of AI in ophthalmology, especially in resource-limited settings.

Main Methods:

• Proposed a novel deep learning architecture, FlexiVarViT, designed for OCT image analysis.
• Incorporated features to handle variable data, such as slice number and resolution, without resizing B-scans.
• Evaluated the model on three diverse datasets from different imaging devices (Spectralis, Optovue) and populations (France, Russia, Iran).

Main Results:

• FlexiVarViT demonstrated high accuracy in detecting and classifying multiple retinal pathologies.
• The model exhibited significant robustness and generalizability across varied patient demographics and imaging systems.
• Achieved superior performance compared to existing state-of-the-art methods.

Conclusions:

• The FlexiVarViT architecture offers enhanced robustness and generalizability for AI-based retinal disease screening.
• The model's ability to process OCT images at native resolution preserves crucial anatomical details.
• FlexiVarViT shows strong potential for widespread clinical application in ophthalmology.