Automatic lung segmentation via a superpixel-based conditional breadth-first search algorithm

Computed Tomography (CT) stands as a preeminent medical imaging modality in computer-aided diagnosis, where accurate segmentation of lung parenchyma from CT slices can assist surgeons to diagnose lung diseases and improve the survival rate and prognostic conditions. This paper presents an automatic algorithm for the lung parenchyma segmentation based on superpixels and graph theory. The superpixels are obtained with simple linear iterative clustering method firstly. Then the conditional breadth first search algorithm is proposed to classify the target superpixels into lung parenchyma and non-lung parenchyma. With a contour refinement algorithm, we get the final segmentation results. Experimental evaluations reveal that the introduced method attains superior segmentation precision and accuracy, evidenced by an average Jaccard Index volume pixel overlap ratio of 96.53% across five distinct types of lung parenchyma image sequences.