Deep Learning Approach for Segmentation and Classification of COVID-19 in Lung CT Scan Images

The COVID-19 pandemic began in 2020 and is still spreading today, this disease has caused many patients to die. COVID-19 is caused by the severe acute respiratory syndrome 2 virus. COVID-19 has been declared a world pandemic by WHO because of the rapid spread and malignancy of this disease. An alternative to detecting COVID-19 is to use a CT scan, However, with the limited number of doctors to detect lesions in COVID-19 patients, an automated system is needed to segment and classify COVID-19 disease. Currently, research into COVID-19 detection using CT scans with deep learning is carried out using single-task learning so that the model can perform segmentation or classification. Some studies use feature extraction from segmentation to train other classification methods that require more time to train the model. The aim of this research is to utilize Swish U net and simplify the classification process. In this research, we proposed a model that can reduce memory usage, speed up the training process, and increase the performance, by using the proposed deep learning model for classification and segmentation with EfficientN et encoding and SwishUnet decoding, we managed to obtain good performance. Proposed model is trained using the dataset of the Italian Society of Medical and Interventional Radiology. For segmentation, sensitivity was obtained with a value of 75.31 % and specificity with a value of 99.44%. Meanwhile, classification succeeded in obtaining a precision performance of 97.72%, recall of 89.58%, and F-measure of 93.47%.