Eye fundus image quality evaluation device and method using transfer learning

Abstract

The present invention relates to an eye fundus image quality evaluation device and method using transfer learning, and more particularly, to an eye fundus image quality evaluation device and method using transfer learning, which evaluate the quality of an eye fundus image after performing data preprocessing on an acquired eye fundus image using transfer learning using a convolutional neural network (CNN) model. Therefore, low-quality images are discarded, only high-quality images are applied to learning, and the convolutional neural network is an image classification artificial intelligence model for learning general images.

Description

technical field [0001] The present invention relates to a fundus image quality evaluation device and method, more specifically, to a fundus image quality evaluation device and method using transfer learning, which utilizes transfer learning using a Convolution Neural Network (CNN) model to After data preprocessing of the obtained fundus images, evaluate the quality of the fundus images, thereby discarding low-quality images, and only applying high-quality images to learning. Among them, the convolutional neural network is an artificial image classification method for learning general images. smart model. Background technique [0002] Diabetic Retinopathy (DR) is a vascular complication of the retina that causes damage to the retina, thereby leading to severe vision loss if not treated promptly. [0003] The risk of blindness due to diabetic retinopathy can be reduced by up to 50% through early detection and appropriate treatment based on regular screening for diabetic retin...

AI Technical Summary

Problems solved by technology

[0012] However, in order to learn the quality evaluation of the fundus image, it is necessary to extract the features of the given complex image by using more than 5 convolution layers and more than 2 fully connected (Fully Connected, FC) layers. Learning process, so there is a problem that the time required for learning in a single CPU environment may consume hundreds to thousands of hours

[0013] Also, when the fundus image is directly used, there is a problem that it takes a lot of time to analyze an unnecessary image area

Images