Analysis of CT scans to differentiate COVID-19 from other lung diseases by using Machin learning and preprocessing

Document Type : Original Article

Authors

Department of Software Engineering, Koya University, Kurdistan Region, Iraq.

10.24271/psr.2024.417244.1393

Abstract

The world’s last pandemic is the coronavirus disease of 2019 (COVID-19), which emerged in Wuhan, China, in December 2019. The World Health Organization (WHO) recognized COVID-19 as a pandemic in March 2020. That is why COVID-19 detection became a very important action for controlling the pandemic. So early, it became the main concern of hundreds of researchers in various aspects. In this study, a detection system of two models of COVID-19 detection is proposed, each composed of feature extraction and classification: the first model adopts a pre-trained ResNet 50 model, and the second model extracts features from the ResNet 50 model and classifies them using SVM. Before feature extraction and classification, the system has a phase of preprocessing. The importance of this step is to prepare the data to get better classification rates. The system recognizes three classes, i.e., COVID-19, normal lung, and other lung diseases. As a reliable public dataset of these three classes is not available, a database was created that includes 735 cases that are collected from Kurdistan hospitals. The system was trained and tested using the created database and gave the following results: The highest accuracy of the first model with three classes of data is 91.1% and 75.7% with and without preprocessing, respectively, and the second model achieved an accuracy of 80.7% and 66.1 with and without preprocessing, respectively. The system was also tested using two classes (normal and abnormal) and gave 93.3% and 92.2% for the first and second models, respectively, with preprocessing.

Keywords

Main Subjects