Lung segmentation in chest radiographs using fully convolutional networks

Abstract

Automated segmentation of medical images that aims at extracting anatomical boundaries is a fundamentalstep in any computer-aided diagnosis (CAD) system. Chest radiographic CAD systems, which are used to detectpulmonary diseases, first segment the lung field to precisely define the region-of-interest from which radiographic patternsare sought. In this paper, a deep learning-based method for segmenting lung fields from chest radiographs has beenproposed. Several modifications in the fully convolutional network, which is used for segmenting natural images to date,have been attempted and evaluated to finally evolve a network fine-tuned for segmenting lung fields. The testing accuracyand overlap of the evolved network are 98.75% and 96.10%, respectively, which exceeds the state-of-the-art results.