Covid-19 Detection Based-On CT-Scan Images Using Inception Deep Learning

The SARS-Cov-2 strain caused COVID-19, inflicting mild to moderate respiratory problems. The spread of COVID-19 is extremely fast which has resulted in the number of victims who have been declared dead to date, up to 2,587,225. There are several ways to reduce the spread of COVID-19, one of which is early detection. Currently, there are alternative methods used for early detection, one of which is the neural network method. Deep learning is one type of artificial neural network that is often used for the detection of several kinds of diseases. In this study, we classify CT-Scan images of the lungs based on two classes, namely CT_COVID and CT-NonCOVID, using two models, Inception-v3 and Inception-v4. The total CT-Scan image data used is 2038 and comes from the Kaggle.com website. Results obtained were then compared with standard performance metrics and then analyzed between the best models among the models used in the COVID-19 classification. From the results of the study, the Inception-v3 model obtained an average accuracy value of 93.96%, a precision value of 90.57%, a recall value of 95.65%, a specificity value of 92.81% and an f-score value of 92.51% and The Inception-v4 model obtained an average accuracy value of 86.41%, a precision value of 77.01%, a recall value of 91.18%, a specificity value of 83.77
^%
and an f-score value of 83.38%. Based on the research results, the method with the best performance in COVID-19 classification is the Inception-v3 model because the Inception-v3 model has more layers, with a total of 48 layers and utilizes the idea of factorization that is more suitable for CT-Scan image classification which has low contrast visualization.