Deep learning based early stage diabetic retinopathy detection using optical coherence tomography

Abstract Diabetic retinopathy (DR) is one of the leading causes of preventable blindness globally. Performing retinal examinations on all diabetic patients is an unmet need, and detection at an early stage can provide better control of the disease. The objective of this study is to provide an optical coherence tomography (OCT) image based diagnostic technology for automated early DR diagnosis, including at both grades 0 and 1. This work can help ophthalmologists with evaluation and treatment, reducing the rate of vision loss, and enabling timely and accurate diagnosis. In this work, we developed and evaluated a novel deep network – OCTD_Net, for early-stage DR detection. While one of the networks extracted features from the original OCT image, the other extracted retinal layer information. The accuracy, sensitivity and specificity was 0.92, 0.90 and 0.95, respectively. Our analysis of retinal layers and the features learned by the proposed network suggests that grade 1 DR patients present with significant changes in the thickness and reflection of certain retinal layers. However, grade 0 DR patients do not have such significant changes. The heatmaps of the trained network also suggest that patients with early DR showed different textures around the myoid and ellipsoid zones, inner nuclear layers, and photoreceptor outer segments, which should all receive dedicated attention for early DR diagnosis.