Automatic Segmentation of COVID-19 Pneumonia Lesions and its Classification from CT images: A Survey

Rathod, S. R.; Khanuja, H. K.

The rapidly spreading coronavirus disease-2019 (COVID-19) has infected over 64 million people in over 200 countries and territories as of December, 2020. To provide appropriate medical treatment to patients and also to safeguard the uninfected individuals, it is very essential to detect COVID-19 in the early stage. For this purpose, a survey is done to identify works carried out to perform automatic diagnosis of COVID-19 disease in chest computed tomography (CT). In particular, focusing on noise robust segmentation approaches to identify infection areas in lungs for taking appropriate decisions in the process of diagnosis. Unstable distributed patterns of the infected areas between the CAP and COVID-19 exist, partially because of fast progress of COVID19 when symptom onset. Our survey results show that advancements in Artificial Intelligence (AI) application such as machine learning and deep learning has been doing remarkable progress by helping radiologists in the diagnostic process of COVID-19. AI helps us answer some critical questions related to region of infection and spread of coronavirus. Some works are carried out with less clean data and more noisy labels. Results reveal that CNN and UNet++ have achieved better accuracy compared to other deep learning classifiers. In future, we propose a Noise Robust Segmentation approach with a recurrent neural network (RNN) and clustering method to concentrate on infection areas in lungs when taking selective diagnostic actions. © 2021 IEEE.