Enhancing COVID-19 diagnosis and severity evaluation through machine learning algorithms applied to ct images
Yogendra Narayan Prajapati, Manish Sharma, Farooque Azam, Chitra Selvan, Arun Biradar
Abstract
There are still many important obstacles that make evaluating COVID-19 as a pandemic difficult. Utilising all of the tools and resources available in the field is essential to overcoming these obstacles. The popular reverse transcription polymerase chain reaction (RT-PCR) technique is useful for detecting viral presence, but it is not very good at determining the severity of a disease. In this work, we present a simplified and reliable automated method that uses CT scan images to classify COVID-19 severity into three stages: mild, moderate, and severe. Three types of characteristics are extracted for this classification: infection ratio, statistical texture features (mean, standard deviation, skewness, and kurtosis), and GLCM/GLRLM texture features. The classification is based on a straightforward segmentation approach. For scan classification, four machine learning methods are used: decision trees, support vector machines, K-nearest neighbours, and Naïve Bayes. Four essential phases are included in our suggested model: preprocessing, feature extraction, classification, and performance assessment. using Support Vector Machines (SVM) yields impressive segmentation accuracy of 99.12% for normal, 98.24% for mild, 98.73% for moderate, and 99.9% for severe COVID-19 infection. The model outperforms previous models with an amazing area under the curve (AUC) of 0.99. This development is expected to improve diagnostic accuracy, facilitate faster evaluation of infection stage, and guarantee prompt, customised therapy delivery by medical practitioners.