Probabilistic Atlas-guided Eigen-organ Method for Simultaneous Bounding Box Estimation of Multiple Organs in Volumetric CT Images
Cong Yao, Takashige Wada, Akinobu Shimizu, Hidefumi Kobatake, Shigeru Nawano · 2006
We propose an approach for the simultaneous bounding box estimation of multiple organs in volumetric CT images. Local eigen-organ spaces are constructed for different types of training organs, and a global eigen-space, which describes the spatial relationships between the organs, is also constructed. Each volume of interest in the abdominal CT image is projected into the local eigen-organ spaces, and several candidate locations are determined. The final selection of the organ locations is made by projecting the set of candidate locations into the global eigen-space. A probabilistic atlas of organs is used to eliminate locations with low probability and to guide the selection of candidate locations. Evaluation by the leave-one-out method using 10 volumetric abdominal CT images showed that the proposed method provided an average accuracy of 80.38 % for 11 different organ types.