The Research On Automatic Annotation Method Of Cataract Fundus Image Based On Semi-supervised Learning

According to the data,some common eye diseases can be detected by the observation of fundus images,and cataract is one of the common eye diseases.This paper mainly uses the image recognition technology in machine learning algorithm to diagnose and identify the fundus image of cataract,and proposes a semi-supervised learning algorithm that can make full use of unlabeled fundus image data for semi-supervised learning,to improve the accuracy of the test dataset.The semi-supervised learning based cataract fundus image automatic classification and labeling system proposed in this paper has important scientific research value and clinical significance for reducing medical costs,alleviating the shortage of doctor resources and improving the diagnosis rate of cataract diseases.This paper has mainly completed the following aspects:In the cataract fundus image preprocessing stage,image graying is used to convert the original OCT colorful fundus image into a grayscale image,and the pixel values of the image are classified into the same grayscale range using pixel value normalization.The image is initially enhanced using the Contrast Limited Adaptive Histogram Equalization(CLAHE)method,and finally uses gamma correction to remove the effects of uneven illumination on the image.In the fundus image feature extraction stage,the color features,texture features,wavelet features,sketch features and vascular information features of the fundus image are extracted.In the semi-supervised algorithm implementation phase,we use two semi-supervised learning algorithms,one is a semi-supervised algorithm based on traditional machine learning algorithms,and the other one is based on deep neural network(GAN).The dataset used in this paper includes 4000 labeled data and 3851 unlabeled data,of which 4000 data sets are classified into 4 categories.During the experiment,4000 images of the labeled cataract eye were divided into training,validation and test sets according to a ratio of 2:1:1.Finally,the semi-supervised learning-based classification model can achieve a maximum of 87.5% of the four classification accuracy on the test set.This result above also fully proves that the classification and labeling of images by computer-aided diagnosis is feasible.To a certain extent,it has promoted the process of putting the cataract automatic classification system into practical medical application.