Hippocampal shape analysis using medial surfaces

Abstract

In magnetic resonance imaging (MRI) research, significant attention has been paid to the analysis of the hippocampus (HC) within the medial temporal lobe because of its importance in memory and learning, and its role in neurodegenerative diseases. Manual segmentation protocols have established a volume decline in the HC in conjunction with Alzheimer's disease, epilepsy, post-traumatic stress disorder, and depression. Furthermore, recent studies have investigated age-related changes of HC volume which show an interaction with gender; in early adulthood, volume reduction of the HC is found in men but not in women. In this paper, we investigated gender differences in normal subjects in young adulthood by employing a shape analysis of the HC using medial surfaces. For each subject, the most prominent medial manifold of the HC was extracted and flattened. The flattened sheets were then registered using both a rigid and a non-rigid alignment technique, and the medial surface radius was expressed as a height function over them. This allowed for an investigation of the association between subject variables and the local width of the HC. With regard to the effects of age and gender, it could be shown that the previously observed gender differences were mostly due to volume loss in males in the lateral areas of the HC head and tail. We suggest that the analysis of HC shape using medial surfaces might thus serve as a complimentary technique to investigate group differences to the established segmentation protocols for volume quantification in MRI.

Introduction

Magnetic resonance imaging (MRI) is a frequently used tool by neuroscientists and clinicians for volumetric analysis of specific structures of the central nervous system (CNS). In neurodegenerative diseases like Alzheimer's, a region that has been especially prone to volume loss is the medial temporal lobe (MTL), and the specific structures contained within the MTL, like the hippocampus (HC) and parahippocampal gyrus (PHG). These structures are also implicated in post-traumatic stress disorder, epilepsy, depression, and normal aging. Finally, the structures of the MTL are frequently investigated for their involvement in acquisition and retention of memory (Corkin et al., 1997, Mori et al., 1997, Schacter and Wagner, 1999), clearly indicating a need for precise analysis of their involvement in these conditions and functions. In MRI, investigation of these structures is often done using manual segmentation protocols, which yield a quantification of the space the structure occupies within the CNS. Although this technique can successfully demonstrate the involvement of the HC and the PHG in neurodegenerative disease, specific disorders, memory function, and normal aging, it is incapable of showing changes within the structure, taking only the total volume into account. Alternative techniques like voxel-based morphometry (VBM) are not based on manual segmentation of specific structures and do not allow significant findings to be interpreted as evidence of volume loss. Taken together, a precise definition of volume loss within structures in conjunction with normal aging, memory functions, or specific disease states is missing, justifying the need for the development of additional analysis tools.

In both clinical and research contexts, quantitative models for the three-dimensional (3D) shape of these structures might allow an alternative approach for the statistical analysis of their distinct characteristics. In fact, there has been significant progress in the development of such methods for use in computational anatomy, as described by several recent articles in Thomson et al. (2004). A first class of methods uses a feature vector, e.g., determined by spherical harmonics or invariant moment representations (Brechbuhler et al., 1995, Mangin et al., 2003), and attempts to discriminate between classes of shapes using clustering techniques. Such methods are usually numerically stable and allow for the computation of relevant statistics. However, the representations are coarse and hence an interpretation of the results in terms of anatomical changes can be difficult. A second class of methods is based on a representation of an object's surface or interior, along with a study of the mechanical deformations required to transform one object into another (Bookstein, 1997, Cootes et al., 1995, Csernansky et al., 1998). This popular technique is very intuitive, but relies on the use of registration techniques which can be difficult to implement. The calculation of significant statistics from the recovered deformation fields also poses a challenge. A third class makes use of medial representations which provide information on an object's reflective symmetries. Unfortunately, these medial models still need to be registered with each other before any statistics can be derived (Bouix et al., 2001, Styner et al., 2003a). In clinical studies, different classes of methods are often combined in order to obtain intuitive and statistically significant results, see for example Styner et al. (2003b). With respect to structures such as the HC, this last class of methods is attractive because it can be shown that the number of medial surface manifolds is small and hence the representation is compact. The most prominent sheet can be used to register individual data sets in an object-centered way, followed by more precise comparisons (Styner and Gerig, 2001).

Medial models have been successfully used in medical image analysis in a number of contexts, see Golland et al. (1999) and Pizer et al. (1999) for some recent applications. Applying these methods in 3D presents an additional challenge because only a small class of computationally reliable algorithms exist. One such class relies on pruning strategies for 3D Voronoi diagrams (Attali et al., 1997, Näf et al., 1996). Styner and Gerig (2001) have recently developed such a framework where medial models developed by Pizer et al. (2003) and Joshi et al. (2001) are used to segment the HC and spherical harmonics along with a coarse-scale sampled medial description are used to represent its shape.

In this paper, we adopt a similar strategy, but use a novel algorithm we have developed for computing medial surfaces. We apply this method to the analysis of a HC data set for which a volume loss with age in young adulthood has been previously observed in normal male subjects but not in normal females (Pruessner et al., 2001). Our hypothesis is that the volume loss is not uniform over the HC, but is rather localized to specific regions. Our goal is to locate these regions of volume loss as well as the regions where gender differences are most prominent, and to quantify the strength of these effects. To do so, we extract the most prominent medial sheet of the HC for each subject and then flatten it. We then register the flattened sheets in a common coordinate frame using both a rigid and a non-rigid alignment technique. This allows us to express the medial surface radius, which indicates the local width of the HC, as a height function over each flattened surface. We then carry out a statistical examination of the relationship between gender and age and local object width. Our main finding is that the previously observed gender differences were mostly due to volume loss in males in the lateral areas of the HC head and tail.