Effective Semantic Segmentation of Lung X-Ray Images using U-Net Architecture

Image segmentation has become a very important and widely used technique in the medical field. Doctors use various state-of-the-art technologies and techniques such as X-Ray, CT-Scan, MRI to detect the abnormalities. We have used lung X-Ray dataset as X-Ray, which remains affordable to most of the people. Here we used a custom loss function to detect the abnormalities in the X-Rays of the lungs. The custom loss function is a combination of the dice loss, inverse dice loss and binary cross-entropy loss. We have used two models namely U-Net and Seg-Net in the process. We have compared custom loss with every other loss and the table in result section shows that the custom loss performs better for Dataset. U-Net performed better for custom loss as compared to SegNet.