Segmentation and Localization of Anatomical Structures in Chest X-Ray Images using Advance Deep Learning Architectures

Abstract

The COVID 19 pandemic has been a very tough time for people around the globe, be it medical professionals who had to work for long hours or patients who had to deal with shortage of medical professionals or delays in their diagnosis. Chest X-rays have been a valuable tool for identifying COVID in a patient and tracking its progression along with other techniques. However, due to the large number of patients and by extension, chest x-rays, the healthcare professionals are facing a real problem. Therefore, any technique that can help in early diagnosis and reduces effort of medical professionals can essentially be lifesaving. Recently, many researchers have tried to help medical professionals by using advanced deep learning techniques such as Convolutional Neural Network for automatic diagnosis from chest X-rays. In this research, use of multiple advance deep learning like Yolact and Yolact++ to segment and localize anatomical structures in chest x-ray image is explored. This would help medical professionals to look at different anatomical structures independently and this would reduce their effort and time consumption in diagnosis. Furthermore, isolation of these anatomical structures can also help train other specific deep networks for diagnosis of diseases as correct localization will help reduce the noise in the images