COVID-19 CT Image Segmentation

COVID-19 was first identified in Wuhan, China. This virus is spreading worldwide, and to date, the number of new infections and their variants is still increasing in many countries. Early detection of this virus is critical to identify whether the patient is infected with COVID-19 or not. So far, RT-PCR is known as the most common method for COVID-19 detection. But compared with lung CT scan, lung CT scan has higher sensitivity, and it is easy to use in COVID-19 infection screening. One of the many steps that are needed in COVID-19 screening is the lung segmentation process. Many segmentation models have been developed by many researchers. And, through this paper, we presented the result of our exploration to the nine segmentation models PyTorch (Unet, Unet++, Manet, Linknet, FPN, PSPNet, PAN, DeepLabV3, and DeepLabV+) and compared each performance and see the result. Based on the experiments, we can conclude that Unet++ performs better than other models. It might be because U-Net++ has a robust network architecture, and it is initially designed for medical image segmentation. U-Net++ is also the largest model that we implemented, which might infer that it has more flexibility because it has more parameters to adjust. It could be the reason for the model to have better performance.