Comparative Visualization of Pelvic Organ Segmentations (DAEV)

Automatic segmentation of pelvic organs plays a major role in prostate cancer treatmentand has high accuracy requirements. Segmentation experts are continuously workingon improving their algorithms. However, natural anatomical variability of organs andstructures is a common reason for which segmentation algorithms fail. Understanding whyan algorithm fails in a specific case is of major importance. Segmentation experts expectthat the shape and size of the organs can play an important role in the performance oftheir algorithms, but current means of exploration and analysis are limited and do not provide the necessary insight. This thesis discusses the design and implementation of a web-based application allowing for easy exploration and analysis of shape variability in order to generate hypotheses about the relation between algorithm performance and shape of organs. A new way of comparatively visualizing multiple organs of multiple patients is introduced for a detailed shape comparison. The application was tested with segmentation meshes of a cohort of 17 patients, each consisting of four pelvic organs and two organ-interfaces, which are labeled and have per-triangle correspondence. The proposed tools already allow to quickly identify mis-segmented organs and hypothesize about the relation of variability to anatomical features as well as segmentation quality. The approach was applied on pelvic organ segmentations, but it can be extended to other applications like comparison of segmentation algorithms or analysis of anatomical variability in general.