Figure 7

Comparison of various graph-based algorithms. a ROI of lung CT images with manual delineations. Demonstrations of the resulting segmentation: b random walks with restart [41], c nonparametric higher-order learning technique [42], and d the proposed algorithm. APHS appears to outperform the other approaches in terms of the quality of results. Note that marking roughly with semantic labels, i.e., lung anatomies (e.g., green parenchyma, blue lung wall, yellow vessel or GGO area, red nodule with solid part, white segment boundaries) is as the same prior information for each algorithm.