Separate Detection of Stromal and Epithelial Corneal Edema on Optical Coherence Tomography Using a Deep Learning Pipeline and Transfer Learning

The accurate detection of corneal edema has become a topic growing interest with the generalization endothelial keratoplasty. Despite recent advances in deep learning for detection, problem minimal remains challenging. Using transfer and limited training set 11 images, we built model to segment epithelium, which is part three-model pipeline detect edema. A second third are used on stroma alone epithelium. validation 233 images from 30 patients consisting three groups (Normal, Minimal Edema important Edema) was compare results our new previous model. mean fraction (EF), defined as number pixels detected divided by total cornea, calculated each image. With model, EF not statistically different between Normal (p = 0.24). current pipeline, higher group compared < 0.01). described constitutes an adjustable framework based optical coherence tomography yields better performances cases or localized