Detecting 3D Corpus Callosum abnormalities in phenylketonuria

Abstract

Phenylketonuria (PKU) is a genetic disorder characterised by an inability to metabolise phenylalanine. Several studies have reported that the Corpus Callosum (CC) is one of the most severely affected structures with respect to volume loss in early treated PKU patients. In this work, we aim to detect the abnormalities of the CC in PKU from both global and local perspectives. 3D models of the CC are extracted from MRI data using a semiautomatic segmentation method. In the global analysis, raw and scaled volumes of the CC are compared between PKU patients and the controls. An oriented bounding box of the CC is constructed and its length, width and height are used as the MRI traits in our study. The raw and scaled values of these traits are compared between patients and controls. In the local analysis, shape differences at every surface point of the CC between PKU patients and the controls are computed using Hotelling T(2) two-sample metric followed by a permutation test. The height of the CC is found to be significantly shorter in the patients and significant shape abnormalities in the genu and splenium of the CC is also found in the patients.

A statistical geometrical description of the human liver for probabilistic occupant models

“…Several correspondence approaches to determine the corresponding landmarks between shapes have been proposed and compared in literature (Chen and Shapiro, 2009;Dalal et al, 2007;He et al, 2009;Lamecker et al, 2002;Styner et al, 2003;van de Giessen et al, 2010). Dalal et al (2007) proposed a correspondence approach, which established landmarks on a selected shape (template) and used the shortest distances between the template landmarks and the points on the target surfaces to determine the corresponding target landmarks.…”

“…However, estimating corresponding points based merely on minimum Euclidean distance was found to be inaccurate in regions with large curvatures (Heimann and Meinzer, 2009). In addition, if the two aligned shapes do not have a certain amount of overlapping, the approach based on shortest distance may fail to establish the correct correspondence (He et al, 2009(He et al, , 2010. To avoid these limitations, several improved correspondence algorithms were proposed using both coordinates and surface normal (Brett and Taylor, 2000;He et al, 2009;van de Giessen et al, 2009).…”

“…In addition, if the two aligned shapes do not have a certain amount of overlapping, the approach based on shortest distance may fail to establish the correct correspondence (He et al, 2009(He et al, , 2010. To avoid these limitations, several improved correspondence algorithms were proposed using both coordinates and surface normal (Brett and Taylor, 2000;He et al, 2009;van de Giessen et al, 2009). The target landmark corresponding to each template landmark was defined as the surface point on a target surface with the smallest Euclidean distance and smallest difference between surface normals (He et al, 2009;van de Giessen et al, 2009).…”

“…To avoid these limitations, several improved correspondence algorithms were proposed using both coordinates and surface normal (Brett and Taylor, 2000;He et al, 2009;van de Giessen et al, 2009). The target landmark corresponding to each template landmark was defined as the surface point on a target surface with the smallest Euclidean distance and smallest difference between surface normals (He et al, 2009;van de Giessen et al, 2009). In addition, it was shown (Belongie et al, 2002) that the iterative thin-plate spline (TPS) algorithm (Bookstein, 1989) could efficiently refine the corresponding landmarks faster than the landmark sliding approach (Dalal et al, 2007) without loss of accuracy.…”

“…In addition, it was shown (Belongie et al, 2002) that the iterative thin-plate spline (TPS) algorithm (Bookstein, 1989) could efficiently refine the corresponding landmarks faster than the landmark sliding approach (Dalal et al, 2007) without loss of accuracy. Therefore, the correspondence approach considering the shortest distance and surface normals (He et al, 2009;van de Giessen et al, 2009) and the iterative TPS algorithm (Belongie et al, 2002) were applied in the current study.…”