A novel method development for classification of COVID-19 Pneumonia in lungs Using Machine learning approach

ABSTRACT

Artificial intelligence may be used to identify COVID-19 pneumonia (also known as pneumococcal meningitis) (AI). AI algorithms are also under scrutiny for their resilience and vulnerability, as are the datasets and research methods used to get the data. AI-driven COVID-19 pneumonia detectors that use our own data from retrospective clinical studies might help overcome these difficulties. In order to assess statistically the research designs, we optimized five deep learning architectures, applied development techniques by altering data distribution and introduced several detection scenarios to test the durability and diagnostic performance of the models. To a greater extent than the present data volume, detection model performance is influenced by hyper parameter adjustment. Sn, sp, and PPV are the three most important metrics in a two-class detection situation, and a method called InceptionV3 has the best of all three. It was shown that models had improved overall performance, with 91-96 percent Sn and 94-98 percent Sp and 91-96 PPV, compared to three-class detection results. Accuracy, F1 scores and g means are all higher than 96% accurate in InceptionV3, according to InceptionV3. For the identification of COVID-19 pneumonia, InceptionV3 had the greatest results, with an AUC of 99. An AUC of 0.98 distinguishes CoVID-19 pneumonia from other kinds of pneumonia, and a micro-average of 0.99 was achieved for the remaining classes.