Postmortem Lung SSM and Lung Segmentation
Unlike postmortem liver, differences in shape of a lung between an in vivo patient and a cadaver are relatively small. Thus, the postmortem SSM was trained by only in vivo lung labels. The shape description ability, however, might be limited, in particular at boundaries neighboring to the liver and heart, because the boundaries are deformed in a cadaver, as explained in Sect.220.127.116.11. The development of a postmortem lung SSM trained from postmortem lung labels and/or synthesized ones is another important goal.
The most difficult problem in postmortem lung segmentation is changes in appearance, or CT values, caused by severe pathology and/or postmortem changes. A prior knowledge of shape and constraints by surrounding organs is necessary to solve the problem. The proposed method in Sect. 4.4.4 employs an SSM to provide a priori knowledge of shape. In addition, extracted body cavity, aorta, as well as liver were used to set constraints on the shape and location of lung. However, the segmentation performance showed in Sect. 18.104.22.168 is not sufficient for practical use. One possible reason is that the spatial extent of pathologies and postmortem changes are highly different among cases, which means that a priori knowledge and constraints necessary for segmentation are on a case-by-case basis. Thus, development of an algorithm to select appropriate constraints adaptively is vital.
Miscellaneous Future Topics
It will be interesting to extend an SSM and a segmentation algorithm to other organs and multiple organs in a cadaver (see Sect. 3.9.6 for in vivo multiple organs). A comparative study of SSMs and segmentation algorithms using more postmortem CT data is also desirable. Modeling both shape and appearance (CT value) changes from an in vivo patient to a cadaver will be essential not only to improve segmentation performance but also to aid in determining the cause of death and to estimate the time after death accurately.