EE-Net: An edge-enhanced deep learning network for jointly identifying corneal micro-layers from optical coherence tomography

Abstract

Accurate delineation of the corneal micro-layers depicted on optical coherence tomography (OCT) plays an important role in computerized detection and diagnosis of various corneal diseases (e.g., keratoconus and dry eye). In this study, we present a novel edge-enhanced convolutional neural network termed EE-Net to automatically delineate three corneal micro-layers from OCT images at the same time, including the epithelium layer, Bowman’s layer, and stroma layer. We innovatively introduced a novel convolutional block and incorporated them with the existing BiO-Net network. Our experiments showed that the developed network achieved a dice similarity coefficient (DSC) of 0.9314, an intersection over union (IOU) of 0.8839, a Matthew's correlation coefficient (MCC) of 0.9314, and a sensitivity of 0.9320 on average respectively for the three different corneal micro-layers, which consistently outperformed available classical networks.