Morphometric Analysis of Hippocampal Shape in Mild Cognitive Impairment: An Imaging Genetics Study Li Shen and Andrew J. Saykin Center for Neuroimaging, Dept. of Radiology Center for Computational Biology & Bioinformatics Indiana University School of Medicine 950 W Walnut St, R2 E124, Indianapolis, IN 46202 Email: shenli@iupui.edu, asaykin@iupui.edu Moo K. Chung Biostatistics & Medical Informatics Univ. of Wisconsin Madison 1300 University Ave., Madison, WI 53706 Email: mkchung@wisc.edu Heng Huang Computer Science & Engineering Univ. of Texas at Arlington Box 19015, 416 Yates St., Arlington, TX 76019 Email: heng@uta.edu Abstract—A computational framework is presented for surface based morphometry to localize shape changes between groups of 3D objects. It employs the spherical harmonic (SPHARM) method for surface modeling and random field theory (RFT) for statistical inference. Several new components are introduced to overcome previous limitations: (1) a general linear model is used to facilitate controlling for covariates; (2) a new SPHARM registration method SHREC is proposed to better align SPHARM models; and (3) an estimated smoothness is used in RFT-based analysis to obtain more accurate results. This framework is applied in a mild cognitive impairment (MCI) study to examine hippocampal shape changes related to diagnostic and genetic conditions. Several interesting findings from our analyses suggest combining imaging phenotypes and genetic profiles has the po- tential to elucidate biological pathways for better understanding MCI and Alzheimer’s disease. I. I NTRODUCTION Statistical morphometric analysis is used in biomedical imaging to study various structures of interest, and aims to identify morphometric abnormalities associated with a partic- ular condition in order to aid diagnosis and treatment. We have previously developed a surface-based morphometry (SBM) framework and applied it to a neuroimaging genetics study for relating hippocampal shape changes to certain conditions in mild cognitive impairment (MCI) [18], [19]. In this work, we introduce several novel components into our SBM framework in order to achieve more accurate and more effective results. We use the same MCI data to demonstrate the effectiveness of our new framework. MCI [14] is characterized by memory complaints and impairment in the absence of dementia and confers a high risk for Alzheimer’s disease (AD). Brain imaging methods for identifying medial temporal morphological abnormalities [4], [16] in circuits required for learning and memory have been studied for early diagnosis and treatment of MCI and AD. The connection between genotype and imaging phenotype has yet to be established, which can help identify possible genetic risk factors for MCI and AD. Although Apolipoprotein E (APOE) appears related to sub- tle cognitive and neuroimaging changes [20], late-onset AD is a complex disorder that undoubtedly involves many genes and polymorphisms. For example, the Interleukin-6 (IL-6) gene is a proinflammatory cytokine involved in neuronal signaling that appears to reduce hippocampal neurogenesis [12], and the single-nucleotide polymorphism (SNP) of IL-6 in the -174 promoter region appears to modulate the reduction of medial temporal volume and gray matter concentration in older adults with memory decline [15]. In this study, we perform morphometric analysis aiming at a global and local quantitative representation of hippocampal shape changes related to certain conditions in MCI. One condition to evaluate is the interaction between morphometric changes of the hippocampus and the IL-6 -174 SNP. We classify this as an imaging genetics study, where imaging genetics [9] refers to the study of genetic variation using imaging measures as phenotypes. To achieve the above goal, we develop an improved SBM framework that can localize regionally specific shape changes between groups of 3D objects. Our framework incorporates spherical harmonic (SPHARM) method [2] for surface mod- eling, heat kernel smoothing [5] for increasing surface signal to noise ratio, and random field theory (RFT) [5], [21] for sta- tistical inference directly on the surface. This new framework overcomes several limitations of our previous method [18], [19], which are described below. Our previous method employs t-test for statistical inference and cannot exclude the effect of any covariate. In this work, we perform statistical analysis using general linear model (GLM), which allows us to obtain more accurate results by removing effects of covariates. For example, to identify hippocampal shape changes in MCI, we often need to remove the age effect. To relate shape changes to IL-6 SNP, the effect of APOE-e4 needs to be excluded. Our previous method uses the first order ellipsoid (FOE) for registering SPHARM models, which may not work in general. In this work, we present a general registration approach called SHREC, based on minimizing the distance between the corresponding SPHARM models [17]. We demonstrate that SHREC can not only create more accurate registration than the FOE approach but also do it efficiently. Smoothness estimation is a key step in RFT-based analysis [21]. The smoothness measure used in our previous study is predicted from heat diffusion equations [5]. A recent study [8]