Analyses of Missing Organs in Abdominal Multi-Organ Segmentation

Current methods for abdominal multi-organ segmentation (MOS) in CT can fail to handle clinical patient population with missing organs due to surgical removal. In order to enable the state-of-the-art atlas-guided MOS for these clinical cases, we propose 1) statistical organ location models of 10 abdominal organs, 2) organ shift models that capture organ shifts due to specific surgical procedures, and 3) data-driven algorithms to detect missing organs by using a normality test of organ centers and a texture difference in intensity entropy. The proposed methods are validated with 34 contrast-enhanced abdominal CT scans, resulting in 80% detection rate at 15% false positive rate for missing organ detection. Additionally, the method allows the detection/segmentation of abdominal organs from difficult diseased cases with missing organs.