An Efficient Variant of Fully-Convolutional Network for Segmenting Lung Fields from Chest Radiographs

Automatic analysis of chest radiographs using computer-aided diagnosis (CAD) systems is pivotal to perform mass screening and detect early signs of various abnormalities in patients. In a chest radiographic CAD system, segmentation of lung fields is a pre-requisite step to precisely define region-of-interest and is subsequently used by other stages of the CAD system. However, automatic segmentation of lung fields is extremely challenging due to substantial variation in lung's shape and size. It still remains an active area of research with sufficient scope to explore new horizon. This paper presents an efficient variant of fully-convolutional network that performs segmentation of lung fields in chest radiographs. The major contribution of this work is in proposing a deep learningbased segmentation architecture that is especially suitable for lung segmentation. The proposed architecture is trained and evaluated on publicly available standard datasets. The architecture achieves the testing accuracy of 98.92% and testing overlap of 95.88% which is better than state-of-the-art methods.