DIAGNOSIS OF COVID-19 THROUGH RADIOLOGY IMAGES USING MACHINE LEARNING TECHNIQUES

Abstract

The period of the world being infected with the corona virus is considered one of the difficult times that cannot be forgotten. The suffering was great in most countries at the beginning of the virus. There were not many tools to confront the virus, which led to many infections and losses. After eliminating the virus and eliminating it thanks to many technologies and developments, it became necessary to study cases of infection and provide a tool for diagnosing those infected. Many datasets were made available to researchers, and it became necessary for us to compare the performance of available systems and develop an accurate and effective model for diagnosing the virus. The importance that we present in our proposed idea is that there is an effective way to identify those infected with the virus. The model's performance can be used to diagnose diseases and other injuries by analyzing ct images. Those infected with the virus are diagnosed with pneumonia and cases are normal. Alexnet, resnet, vgg16, lenet, googlenet and zfnet are effectively used in diagnosis and classification. The data we obtained from al zahra hospital was specifically adapted for use in our proposed system. The total number of labeled samples is 6,396 images, divided into 4,237 pneumonia, 1,583 normal, and 576 infected with covid-19. Ten-fold cross-validation is commonly used to evaluate the accuracy of algorithms. This involves dividing the data set into two subsets: the training set, where the classifier is built, and the testing set, where the classifier is evaluated. The accuracy of the algorithm is estimated by averaging the value of the results obtained from the ten-fold cross-validation procedure. To achieve this, the dataset is evaluated for our results, and thus our proposed system features accurate diagnosis and classification. Class loading based on image dataset (labels: covid, pneumonia, normal). Implement cnn with transfer learning in two ways: • load the pre-trained alexnet network, and fine-tune the last three layers for the new classification by training the network on the training data. • reuse the pre-trained alexnet network to extract features from the “fc7” layer and then classify those features using a support vector machine (svm) model where those features are extracted from the training and testing images. The classification is evaluated using test examples and examining the difference between images using cnn as classifier and feature extraction. Month year, …….page (example: november 2017, 129 pages).