Tournament Based Ranking CNN for the Cataract grading

摘要

An object of the present invention is to provide a model capable of effectively determining the grade of cataract by using a very deep artificial neural network. To this end, provided are a method for operating a tournament-based ranking convolutional neural network for cataract diagnosis and a device thereof. The method for operating a tournament-based ranking convolutional neural network for cataract diagnosis provided in the present invention comprises: a step of dividing a plurality of classes constituting a cataract dataset into a plurality of sets; a step of repeating the process of dividing all the classes into the plurality of sets until one class belongs to one set; a step of generating a tournament structure in the order divided into the plurality of sets; a step of training a binary convolutional neural network (cnn) model for determining the cataract dataset divided into the plurality of sets; and a step of obtaining a binary output of a binary convolutional neural network model of an input image by using the trained binary convolutional neural network model and determining one of a plurality of labels according to the tournament structure based on the obtained binary output.