Multiple Abdominal Organs

Introduction In the human body, various anatomical structures are interrelated in a complex manner. The abdomen is the most appropriate domain to address the problem of representing multi-organ interrelationships. The pioneering work on abdominal multi-organ segmentation by Shimizu et al. was published in 2007 [256]. Since then, several methods have been proposed [180, 223-225,299]. In this subsection, statistical modeling of interrelated organs is addressed, and one of the latest works is described [225]. The organs to be segmented are the liver, spleen, left and right kidneys, pancreas, gallbladder, aorta, and inferior vena cava.

Basic Unit for Modeling Multiple Organs: Prediction-Based CA Models We

begin with two interrelated organs before addressing multiple organs generally. As a typical example of two organs, we first consider the liver and gallbladder. These two organs are closely situated, and the gallbladder handles the bile secreted by the liver. It is desirable to represent interrelations between them in addition to the two separate CA models. To represent two organs statistically, two approaches will be considered, that is, joint probability representation and conditional representation. If we consider joint representation, one possible method is to regard the two organs as one object. One SSM of the two organs can be represented in a hierarchical manner similar to H-SSMs described in the previous section. With conditional representation, the liver can be segmented in a sufficiently accurate manner by a single-organ segmentation method, while the error in gallbladder segmentation is typically much larger than the liver due to its small size and locational variability. Therefore, it would be useful for CA models of the gallbladder to be conditionally modeled under the assumption that the segmentation result of the liver is given. Figure 3.88a shows a conventional as well as a conditional probabilistic atlas of the gallbladder when the liver shape is given, which represents the ambiguity of its shape and location remaining after their prediction from the segmented liver shape. The prediction scheme is formulated using partial least squares regression (PLSR) of the target organ (gallbladder) from the predictor organ, the liver. In this form, the conditional probabilistic atlas is considered to represent the prediction error of PLSR. We call it a prediction-based probabilistic atlas. The prediction error is also represented as an SSM, which is a different form of conditional modeling and called a prediction-based SSM. This form is applied to various interrelated organs. Figure 3.88b shows the prediction-based probabilistic atlas of the pancreas given the liver and spleen shapes, in comparison with a conventional probabilistic atlas.

 
Source
< Prev   CONTENTS   Source   Next >