Mathematical morphology based lung segmentation using multiscale dense pyramid network architecture

Chest radiographs (CXR) are widely used technique in healthcare for the diagnosis and treatment of patients. The recognition of visual irregularities from CXR is most difficult task for expert radiologists as the chest area composed of many sensitive anatomical structures. In this paper, a reliable and secure fully automated mathematical morphology-based architecture (M2LS-Net) is designed for extracting lungs region from CXRs. This method uses dense and multiscale features to segment lungs from chest followed by the use of morphological features to eliminate the non-lung regions to enhance the accuracy of the segmentation. The designed method has attained better performance in comparison to established architectures when evaluated against different performance metrics