ShapeShape versus Size: Improved understanding of the Morphology of Brain Structures 1,2Guido Gerig, 1Martin Styner, 3Martha E. Shenton, 2Jeffrey A. Lieberman 1Department of Computer Science, UNC, Chapel Hill, NC 27599, USA 2Department of Psychiatry, UNC, Chapel Hill, NC 27599, USA 3Department of Psychiatry, VAMC-Brockton, Harvard Medical School, Boston, USA [email protected], [email protected]. Standard practice in quantitative structural neuroimaging is a segmentation into brain tissue, subcortical structures, fluid space and lesions followed by volume calculations of gross structures. On the other hand, it is evident that object characterization by size does only capture one of multiple aspects of a full structural characterization. Desirable parameters are local and global parameters like length, elongation, bending, width, complexity, bumpiness and many more. In neuroimaging research there is increasing evidence that shape analysis of brain structures provides new information which is not available by conventional volumetric measurements. This motivates development of novel morphometric analysis techniques answering clinical research questions which have been asked for a long time but which remained unanswered due to the lack of appropriate measurement tools. Challenges are the choice of biologically meaningful shape representations, robustness to noise and small perturbations, and the ability to capture the shape properties of populations that represent natural biological shape variation. This paper describes experiments with two different shape representation schemes, a fine-scale, global surface characterization using spherical harmonics, and a coarsely sampled medial representation (3D skeleton). Driving applications are the detection of group differences of amhygdala-hippocampal shapes in schizophrenia and the analysis of ventricular shape similarity in a mono/dizygotic twin study. The results clearly demonstrate that shape captures information on structural similarity or difference which is not accessible by volume analysis. Improved global and local structure characterization as proposed herein might help to explain pathological changes in neurodevelopment/neurodegeneration in terms of their biological meaning. 1. Introduction In-vivo imaging studies of brain structures have provided valuable information about the nature of neuropsychiatric disorders including neurodegenerative diseases and/or disorders of abnormal neurodevelopment. Deformities in brain structure in Alzheimer’s and Huntington’s disease are believed to be due the effects of the disease process in adulthood after a period of normal neurodevelopment, while diseases like Autism and Fragile X MICCAI 2001, LNCS 2208: 24-32syndrome are thought to involve abnormal neurodevelopment which give rise to the symptoms of the illness. Schizophrenia, on the other hand, is often subject to conflicting hypotheses about the cause and temporal evolution of the neuropathologic features of the disorder. Structural imaging studies so far have most often focused on volumetric assessment of brain structures, for example full brain or hemispheric gray and white matter, ventricular volume and hippocampus. Increasing evidence for structural changes in small subregions and parts of structures drive development of new structure analysis techniques. Wang [Wang 2000] found that while hippocampal volume did not discriminate schizophrenia groups from control groups, shape measurements did provide a distinct group separation. The paper further discusses that summary comparisons of whole structures ignores the possibility of detecting regional differences. Csernansky [Csernansky et al., 1998] suggests that a full characterization of neuroanatomical abnormalities will increase our understanding of etiology, pathogenesis, and pathophysiology of schizophrenia. Results show that the analysis of hippocampal shape discriminates schizophrenia and control subjects with greater power than volumetry [see also Haller et al., 1996,1997]. [Suddath et al., 1990] found smaller anterior hippocampi in affected vs. unaffected MZ twins. All these studies advocate new morphometry techniques to study shape rather than gross volume and to provide quantitative measures that are not only statistical significant but also neuroanatomically relevant and intuitive. 2. Shape Modeling Surface-based shape representation: We applied a technique for surface parametrization that uses expansion into a series of spherical harmonics (SPHARM) [Brechbuehler, 1995, Szekely et al., 1996, Kelemen, 1999]. The development parallels the seminal work of Cootes & Taylor [Cootes et al., 1995] on active shape models but is based on a parametric object description (inspired by Staib and Duncan [Staib, 1996]) rather than a point distribution model. SPHARM is a global parametrization method, i.e. small local changes can affect all parameters. It allows simple alignment of structures and gives a good initial point-to-point correspondence. Medially-based surface shape representation: As an alternative to surface-based global shape representation, the UNC research group is working on a 3D skeletal representation with coarse to fine sampling (Pizer, 1999, Yushkevich, 2001, Styner, 2000, 2001). The medial shape representation provides locality of width and bending on a hierarchy of scales. Parameters derived from the medial representation are more intuitive than Fourier coefficients and will help to develop shape descriptions expressed in natural language terms.3. Applications in Neuroimaging The following subsections describe applications of surface-based and medially-based shape representation methods in two clinical studies. 3.1 Statistical analysis of amygdala-hippocampal asymmetry in schizophrenia. We studied the asymmetry of the hippocampal complex for a group of 15 controls and 15 schizophrenics (collaboration with M. E. Shenton, Harvard). Asymmetry was assessed by segmentation using deformable models (Kelemen, 1999), by flipping one object across the midsagittal plane, by aligning the reference and the mirrored object using the coordinates of the first ellipsoid, and by calculating the MSD between the two surfaces (Fig.1). Fig. 1. Analysis of amygdala-hippocampal left/right asymmetry. The left hippocampal complex is mirrored, aligned and overlaid (right) to calculate the mean square distance between surfaces. As shape difference measures can be largely influenced