Automatic lung segmentation in HRCT images with diffuse parenchymal lung disease using graph-cut

High-resolution computed tomography (HRCT) is a dedicated medical imaging technique for diffuse parenchymal lung disease evaluation. Such diseases give rise to variability in visual interpretation, leading to the need for computer-aided diagnosis (CAD) systems, for which lung segmentation is a necessary initial step. The developed algorithm is based on the graph-cut technique, which uses an initialization mask produced automatically based on threshold and morphological techniques. Eleven HRCT patient scans were used in this retrospective study. Performance was assessed using ground truth lung segmentation data. Accurate results with surface overlap of 97.42% and an average distance error of 0.92 mm were produced. The main limitation was the difficulty of segmenting lungs with extremely severe disease patterns, while over- and under-segmentation arose only in a limited number of images. The developed method is a good candidate for the first stage of a CAD system for diffuse lung diseases.