Detection of Pathological Myopia and Optic Disc Segmentation with Deep Convolutional Neural Networks

Pathological Myopia (PM) is one of the leading cause of blindness all over the world and is particularly prevalent in Asia. Pathological myopia is a severe case of myopia which is accompanied by degenerative changes in the retina and can lead to non recoverable vision loss. The accurate diagnosis of PM will enable timely intervention and better treatment planning. Current methods for the detection of pathological myopia are manual and subjective. In this study, we have developed a Convolutional Neural Network (CNN) based system to estimate probability/risk of the patient to be diagnosed with PM from his retinal images. We have also segmented optic disc from the retinal images with novel Residual UNet architecture. Validation data results are evaluated from the online server and achieved 0.9973 and 0.901 for the task of PM classification and optic disc segmentation respectively.