Automatic segmentation of juxta-pleural tumors from CT images based on morphological feature analysis

Abstract

Extraction of lung tumors is a fundamental step for further quantitative analysis of the tumor, but is challenging for juxta-pleural tumors due to the adhesion to the pleurae. An automatic algorithm for segmentation of juxta-pleural tumors based on the analysis of the geometric and morphological features was proposed. Initially, the lung is extracted by means of thresholding using 2D Otsu's method. Next a center point is suggested to find a starting point and endpoint of outward facing pleura. A model based on the variation of incline angle was adopted to identify potentially affected regions, and to full segment juxta-pleural tumors. The results were compared with the manual segmentation by two radiologists. Averaged for ten experimental datasets, the accuracy calculated by Dice index between the results of the algorithm and by the two radiologists is 91.2%. It indicates the proposed method has comparable accuracy with the experts (the inter-observer variability is 92.4%), but requests much less manual interactions. The proposed algorithm can be used for segmenting juxta-pleural tumors from CT images, and help improve the diagnosis, pre-operative planning and therapy response evaluation.