OCT Segmentation using Convolutional Neural Network

OCT of retina indicates presence of macular edema, variation in thickness of different layers of retina and choroid. The size of the edema and the thickness of the choroid layers can be ascertained by proper segmentation of the OCT images. In this work, we propose a model using Convolutional Neural Network (CNN) similar to the SegNet architecture for segmenting choroid layers and edema in OCT images. Our CNN model is an encoder - decoder architecture designed specifically for pixel wise classification of images where boundary delineation is vital. To this end, first we train the CNN to obtain the pixel wise label of the choroid and edema. Once these labels are obtained, the pixel labels are converted into binary segmentation using morphological operations followed by edge detection. The results show good consistency when compared with the ophthalmologist's labeling. The idea can be extended to develop a tool to detect retinal disorders automatically.