Detection of multiple retinal diseases in ultra-widefield fundus images using deep learning: data-driven identification of relevant regions

TL;DR

This study develops a deep learning model capable of detecting multiple retinal diseases in ultra-widefield images under realistic conditions, and uses explainability methods to identify key retinal regions relevant for diagnosis.

Contribution

The paper introduces a novel DL model for multi-disease detection in UWF images under realistic scenarios and systematically identifies relevant retinal regions using explainability techniques.

Findings

Model achieves AUC of 0.9206 internally and 0.9841 externally.

Model attends to expected disease regions, especially the posterior pole.

10% of the posterior pole region suffices for comparable performance.

Abstract

Ultra-widefield (UWF) imaging is a promising modality that captures a larger retinal field of view compared to traditional fundus photography. Previous studies showed that deep learning (DL) models are effective for detecting retinal disease in UWF images, but primarily considered individual diseases under less-than-realistic conditions (excluding images with other diseases, artefacts, comorbidities, or borderline cases; and balancing healthy and diseased images) and did not systematically investigate which regions of the UWF images are relevant for disease detection. We first improve on the state of the field by proposing a DL model that can recognise multiple retinal diseases under more realistic conditions. We then use global explainability methods to identify which regions of the UWF images the model generally attends to. Our model performs very well, separating between healthy and diseased retinas with an area under the curve (AUC) of 0.9206 on an internal test set, and an AUC of 0.9841 on a challenging, external test set. When diagnosing specific diseases, the model attends to regions where we would expect those diseases to occur. We further identify the posterior pole as the most important region in a purely data-driven fashion. Surprisingly, 10% of the image around the posterior pole is sufficient for achieving comparable performance to having the full images available.