Postmortem Lung Segmentation
Lung segmentation from a postmortem thoracic CT volume is a challenging task owing to large differences not only in shape but also in appearance from a healthy living lung. The change in appearance from in vivo lung to cadaveric one was caused by both postmortem changes and severe pathologies, such as multiple tumors or a large pleural effusion (see Figs. 4.33 and 4.34). One of the state-of-the-art in vivo lung segmentation algorithms is the multi-shape graph cuts with neighbor constraints . It was reported that the algorithm achieved higher accuracy than conventional one, in particular when applying to lung with atypical shape and pathologies. Such in vivo lung segmentation algorithm might still be effective in segmentation of postmortem lung to an extent. Figure 4.33 presents an example of segmentation results for a case with moderate changes in appearance by the previous segmentation algorithm . It was confirmed that the lung boundaries are extracted successfully by the method (JI: 0.963 [left], 0.830 [right]). It might, however, not succeed in extracting a postmortem lung with more severe deformation and/or changes in appearance, as in Fig. 4.34.
This subsection presents a lung segmentation algorithm for a postmortem thoracic CT volume . It is a modified version of the in vivo lung segmentation algorithm . There is twofold contribution: First, to deal with a large difference in appearance of lung that leads to failure in rough location estimation of lung, a robust location estimation algorithm is proposed. It uses the result of postmortem liver segmentation explained in the previous section . Second, a new adaptive
Fig. 4.33 Segmentation results of moderate cases. The previous segmentation algorithm  succeeded in extracting the boundaries
Fig. 4.34 Segmentation results of severe cases with estimated top and bottom sections of the lung. While the previous algorithm significantly failed in segmentation, the proposed algorithm improved the lung boundaries (Revised version of figure 1 in )
weight between energy terms in the graph cuts is presented. It adaptively balances between different terms according to the reliability of the rough lung segmentation result. The results of applying the algorithm to 32 postmortem thoracic CT volumes are presented in this subsection.