Josef Pelikán

Technical Note: Geometric Morphometrics and Sexual Dimorphism of the Greater Sciatic Notch in Adults From Two Skeletal Collections: The Accuracy and Reliability of Sex Classification

The greater sciatic notch (GSN) is one of the most important and frequently used characteristics for determining the sex of skeletons, but objective assessment of this characteristic is not without its difficulties. We tested the robustness of GSN sex classification on the basis of geometric morphometrics (GM) and support vector machines (SVM), using two different population samples. Using photographs, the shape of the GSN in 229 samples from two assemblages (documented collections of a Euroamerican population from the Maxwell Museum, University of New Mexico, and a Hispanic population from Universidad Nacional Autónoma de México, Mexico City) was segmented automatically and evaluated using six curve representations. The optimal dimensionality for each representation was determined by finding the best sex classification. The classification accuracy of the six curve representations in our study was similar but the highest and concurrently homologous cross-validated accuracy of 92% was achieved for a pooled sample using Fourier coefficient and Legendre polynomial methods. The success rate of our classification was influenced by the number of semilandmarks or coefficients and was only slightly affected by GSN marginal point positions. The intrapopulation variability of the female GSN shape was significantly lower compared with the male variability, possibly as a consequence of the intense selection pressure associated with reproduction. Males were misclassified more often than females. Our results show that by using a suitable GSN curve representation, a GM approach, and SVM analysis, it is possible to obtain a robust separation between the sexes that is stable for a multipopulation sample.

Low-rank matrix approximations for Coherent point drift

We test non-rigid Coherent point drift with different low-rank matrix approximations.We designed a clustering-based method of approximating eigenvectors.CPD preprocessing times have been decreased to about one third.Both precision of registration and robustness are sufficient for most applications. Coherent point drift (CPD) is a powerful non-rigid point cloud registration algorithm. A speed-up technique that allows it to operate on large sets in reasonable time, however depends on efficient low-rank decomposition of a large affinity matrix. The originally used algorithm for finding eigenvectors in this case is based on Arnoldis iteration which, though very precise, requires the calculation of numerous large matrix-vector products, which even with further speed-up techniques is computationally intensive. We use a different method of finding that approximation, based on Nystrom sampling and design a modification that significantly accelerates the preprocessing stage of CPD. We test our modifications on a variety of situations, including different point counts, added Gaussian noise, outliers and deformation of the registered clouds. The results indicate that using our proposed approximation technique the desirable qualities of CPD such as robustness and precision are only minimally affected, while the preprocessing times are lowered considerably.

Semiautomatic extraction of cortical thickness and diaphyseal curvature from CT scans

The understanding of locomotor patterns, activity schemes, and biological variations has been enhanced by the study of the geometrical properties and cortical bone thickness of the long bones measured using CT scan cross-sections. With the development of scanning procedures, the internal architecture of the long bones can be explored along the entire diaphysis. Recently, several methods that map cortical thickness along the whole femoral diaphysis have been developed. Precise homology is vital for statistical examination of the data; however, the repeatability of these methods is unknown and some do not account for the curvature of the bones. We have designed a semiautomatic workflow that improves the morphometric analysis of cortical thickness, including robust data acquisition with minimal user interaction and considering the bone curvature. The proposed algorithm also performs automatic landmark refinement and rigid registration on the extracted morphometric maps of the cortical thickness. Because our algorithm automatically reslices the diaphysis into 100 cross-sections along the medial axis and uses an adaptive thresholding method, it is usable on CT scans that contain soft tissues as well as on bones that have not been oriented specifically prior to scanning. Our approach exhibits considerable robustness to error in user-supplied landmarks, suppresses distortion caused by the curvature of the bones, and calculates the curvature of the medial axis.

Statistical Mesh Shape Analysis with Nonlandmark Nonrigid Registration

The analysis of shape represented as surface meshes is an important tool in anthropology and biomedicine for the study of aging, post-treatment development or sexual dimorphism. Most approaches rely on nonrigid registration using manually placed homologous landmarks, it is however often the case that some regions cannot be landmarked due to the lack of clear anatomical features. We therefore present a method of analyzing and visualizing the variability of a set of surface models that does not rely on landmarks for feature matching and uses coherent point drift (CPD), a nonrigid registration algorithm, instead. Our approach is based on the topology transfer of one arbitrarily selected base mesh to all other meshes with the use of CPD. The procedure ensures the identical meanings of corresponding vertices across the sample and allows the use of multivariate statistics even with shapes that would be difficult to process with methods that rely on landmarks for feature-matching.

Comparing Endocranial Surfaces: Mesh Superimposition and Coherent Point Drift Registration

The endocranial cavity is a major source of information for the assessment of brain morphology in extinct species. Digital molds of the endocranium can be reconstructed from fossil remains. In paleoneurology, these so-called endocasts are examined using shape analysis and multivariate statistical methods to quantify differences among species and individuals. These surfaces are relatively smooth and offer few landmarks; as such, morphometric comparisons are not straightforward, and correspondence search algorithms are necessary to identify loci of equivalent anatomical variation. Many solutions to this so-called correspondence problem have been proposed, but these often require considerable manual input. Here, we present the application in paleoneurology of a correspondence search and symmetrization algorithm originally designed for facial and palatal scans. Homologous representations of surfaces were used to render a computed visualization of differences in shape between modern humans, Neanderthals, archaic humans, and chimpanzees.

Low Level Statistical Models for Initialization of Interactive 2D/3D Segmentation Algorithms

In this paper we present two models which are suitable for interactive segmentation algorithms to decrease amount of user work. Models are used during initialization step and do not increase complexity of segmentation algorithms. Model describe spatial distribution of image values and classification as either foreground or background. Second part of the model is vector field which constrains direction of boundary normals. We show how to use these models in parametric snakes/surfaces framework and minimal graph-cut based segmentation.

An Automatic Algorithm for Tracking Small Intestine in CT Enterography

In this paper we present an automatic algorithm to segment and track small intestine from CT enterography. The algorithm can handle noisy thin-slice data and is adaptable to the greatly varying spatial structure of the organ. Our approach automatically segments all well-distended parts and performs tracking of the intestinal path. Pre-filtered data are segmented with watershed segmentation and then a kNN-based probability function enhances whole parts of the lumen. Post-process based on a robust form of region growing is then used for path tracking.

Towards a GPU-only rod-based hair animation system

Hair is one of the key aspects which make virtual characters believable, with physically based simulation the preferred animation approach. In [Kmoch et al. 2009], we presented a method based on Discrete Elastic Rods [Bergou et al. 2008]. Each guide strand is an elastic rod, given by its centreline, discretized into a chain of n + 2 nodes connected by n + 1 segments, and its material frame, two orthonormal vectors perpendicular to each segment (Fig. 1, left).