Classification of Fundus Images with Pre-Trained Model Using Principal Component Analysis and Support Vector Deviation

Abstract

A new field of study that uses imaging technologies to diagnose diseases is called automated analysis of medical images. Diabetic patients are identified with diabetic retinopathy (DR), a retinal condition. Fundus images taken from suspicious individuals are often used to classify diabetic retinopathy using deep neural networks (DNNs). Through the use of a Gaussian mixture model, and (VGGNet) visual geometry group network, also using singular value decomposition (SVD), in addition, principle component analysis (PCA), Softmax for region segmentation, high dimensional feature extraction, feature selection, and fundus image classification, respectively, the proposed DR classification system achieves a symmetrically optimized solution. 32,228 images from the website KAGGLE dataset were used in the experiments. In terms of classification accuracy and computational time, the suggested VGG-19 DNN-depending DR model fared better than AlexNet and spatial invariant feature transform (SIFT). The classification accuracies of 93.33%, 98.22%, 97.98%, and 98% for FC7-PCA, FC7-SVD, FC8-PCA, and FC8-SVD, respectively, were obtained by using PCA as well as SVD feature selection using fully connected (FC) layers.