STUDYING VENTRICULAR ABNORMALITIES IN MILD COGNITIVE IMPAIRMENT WITH HYPERBOLIC RICCI FLOW AND TENSOR-BASED MORPHOMETRY Jie Shi MS1, Cynthia M. Stonnington MD2, Paul M. Thompson PhD3, Kewei Chen PhD4, Boris Gutman PhD3, Cole Reschke BS4, Leslie C. Baxter PhD5, Eric M. Reiman MD4, Richard J. Caselli MD6, Yalin Wang PhD1, for the Alzheimer’s Disease Neuroimaging Initiative* 1School of Computing, Informatics, and Decision Systems Engineering, Arizona State University, Tempe, AZ, USA 2Department of Psychiatry and Psychology, Mayo Clinic Arizona, Scottsdale, AZ, USA 3Imaging Genetics Center, Institute for Neuroimaging and Informatics, University of Southern California, Los Angeles, CA, USA 4Banner Alzheimer’s Institute and Banner Good Samaritan PET Center, Phoenix, AZ, USA 5Human Brain Imaging Laboratory, Barrow Neurological Institute, Phoenix, AZ, USA 6Department of Neurology, Mayo Clinic Arizona, Scottsdale, AZ, USA Submitted to NeuroImage Please address correspondence to: Dr. Yalin Wang School of Computing, Informatics, and Decision Systems Engineering Arizona State University P.O. Box 878809 Tempe, AZ 85287 USA Phone: (480) 965-6871 Fax: (480) 965-2751 E-mail: [email protected] *Acknowledgments and Author Contributions: Data used in preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). As such, the investigators within the ADNI contributed to the design and implementation of ADNI and/or provided data but did not participate in analysis or writing of this report. A complete listing of ADNI investigators can be found at: http://adni.loni.usc.edu/wp-content/uploads/how_to_apply/ADNI_Acknowledgement_List.pdf. ABSTRACT (260 words) Mild Cognitive Impairment (MCI) is a transitional stage between normal aging and dementia and people with MCI are at high risk of progression to dementia. MCI is attracting increasing attention, as it offers an opportunity to target the disease process during an early symptomatic stage. Structural magnetic resonance imaging (MRI) measures have been the mainstay of Alzheimer’s disease (AD) imaging research, however, ventricular morphometry analysis remains challenging because of its complicated topological structure. Here we describe a novel ventricular morphometry system based on the hyperbolic Ricci flow method and tensor-based morphometry (TBM) statistics. Unlike prior ventricular surface parameterization methods, hyperbolic conformal parameterization is angle-preserving and does not have any singularities. Our system generates a one-to-one diffeomorphic mapping between ventricular surfaces with consistent boundary matching conditions. The TBM statistics encode a great deal of surface deformation information that could be inaccessible or overlooked by other methods. We applied our system to the baseline MRI scans of a set of MCI subjects from the Alzheimer’s Disease Neuroimaging Initiative (ADNI: 71 MCI converters vs. 62 MCI stable). Although the combined ventricular area and volume features did not differ between the two groups, our fine-grained surface analysis revealed significant differences in the ventricular regions close to the temporal lobe and posterior cingulate, structures that are affected early in AD. Significant correlations were also detected between ventricular morphometry, neuropsychological measures, and a previously described imaging index based on fluorodeoxyglucose positron emission tomography (FDG-PET) scans. This novel ventricular morphometry method may offer a new and more sensitive approach to study preclinical and early symptomatic stage AD. Keywords: Alzheimer’s disease, mild cognitive impairment, hyperbolic Ricci flow, tensor-based morphometry 1. INTRODUCTION Mild Cognitive Impairment (MCI) describes individuals who cognitively lie between normal aging and dementia, and MCI often progresses to Alzheimer’s disease (AD). The prevalence of MCI is as high as 19% in persons over age 65, and 29% in those over age 85 (Small et al., 2009). Also, a significant portion of the population diagnosed with life limiting illnesses (LLI) has therapeutic exposures that place them at risk for MCI. As MCI has the greatest risk of progression to dementia, it is a primary focus of interest in aging studies (Roberts et al., 2010). The International Classification of Diseases now has a billing code for MCI (331.83) (ICD, 2010). Over half of those with MCI progress to dementia within 5 years (Gauthier et al., 2006), and early detection of those MCI individuals who will convert to dementia can facilitate earlier intervention, and guide recruitment for clinical trials. Current therapeutic failures in patients with symptomatic memory loss might reflect intervention that is too late, or targets that are secondary effects and less relevant to disease initiation and early progression (Hyman, 2011). As the paradigm in AD research shifts to a new stage, targeting earlier intervention and prevention (Caselli and Reiman, 2013; Langbaum et al., 2013), there is a requirement for biologically grounded, highly objective biomarkers that can help to identify those high AD risk MCI individuals for whom early intervention may be most appropriate. Various neuroimaging techniques can track disease progression and therapeutic efficacy in MCI (Caroli et al., 2012; Chen et al., 2011; Matsuda, 2007b; Petersen, 2011; Petersen and Jack, 2009; Pihlajamaki et al., 2009; Small et al., 2006; Wolf et al., 2003) and others are beginning to identify abnormal anatomical or functional patterns and their rates of decline. Structural magnetic resonance imaging (MRI) has been the mainstay of AD imaging research and has included evaluations of whole-brain (Chen et al., 2007; Fox et al., 1999; Stonnington et al., 2010), entorhinal cortex (Cardenas et al., 2011), hippocampus (Apostolova et al., 2010b; den Heijer et al., 2010; Jack et al., 2003; Jack et al., 2010; Reiman et al., 1998; Thompson et al., 2004a; Wolz et al., 2010), and temporal lobe volumes (Hua et al., 2010), as well as ventricular enlargement (Jack et al., 2003; Jack et al., 2008; Thompson et al., 2004a; Wang et al., 2011). These correlate closely with differences and changes in cognitive performance, supporting their validity as markers of disease progression. Among these biomarkers, ventricular enlargement is a very important measure of AD progression (Frisoni et al., 2010). Owing to the high contrast between the CSF and surrounding brain tissue on T1-weighted images, lateral ventricles can be measured more reliably than hippocampus or other structures, whose boundaries are difficult for experts to agree on (Chou et al., 2010). Furthermore, lateral ventricles span a large area within the cerebral hemispheres and abut several structures relevant to AD including the hippocampus, amygdala and posterior cingulate. Changes in ventricular morphology, such as enlargement, often reflect atrophy of the surrounding cerebral hemisphere which itself may be regionally differentiated (for example, frontotemporal in contrast to posterior cortical atrophy). Regional differences in cerebral atrophy may be reflected in specific patterns of change in ventricular morphology, so accurate analysis of ventricular morphology has the potential to both sensitively and specifically characterize a neurodegenerative process. Many brain imaging based AD studies examine cortical and subcortical volumes (den Heijer et al., 2010; Dewey et al., 2010; Holland et al., 2009; Jack et al., 2004; Jack et al., 2003; Ridha et al., 2008; Vemuri et al., 2008a; Vemuri et al., 2008b; Wolz et al., 2010), but recent research (Apostolova et al., 2010a; Apostolova et al., 2010b; Chou et al., 2009a; Costafreda et al., 2011; Ferrarini et al., 2008b; Madsen et al., 2010; McEvoy et al., 2009; Morra et al., 2009; Qiu et al., 2010; Styner et al., 2005; Tang et al., 2014; Thompson et al., 2004a) has demonstrated that surface-based analyses can offer advantages over volume measures, due to their sub-voxel accuracy and the capability of detecting subtle subregional changes. Higher-order correspondences between brain surfaces are often required to be established in order to statistically compare or combine surface data obtained from different people, or at different time-points. Usually, brain surface registration is done by first mapping the surfaces to be matched onto one common canonical parameter domain, such as a sphere (Fischl et al., 1999b; Thompson et al., 2004b), or a planar rectangle (Shi et al., 2013a), and then registering the surfaces in the simpler parameter domain. The one-to-one correspondences obtained in the parameter domain induce the registration of the 3D brain surfaces. However, it is challenging to apply this framework to ventricular surfaces, due to their concave shape, complex branching topology and extreme narrowness of the inferior and occipital horns. Thus, surface-based subregional analysis of ventricular enlargement is notoriously difficult to assess, exemplified by the conflicting findings regarding genetic influences on ventricular volumes (Chou et al., 2009b; Kremen et al., 2012). Pioneering ventricular morphometry work (Paniagua et al., 2013; Styner et al., 2005) used spherical harmonics to analyze ventricular surfaces where each ventricular surface was mapped to a sphere and registered to a common template. However, as demonstrated previously (Wang et al., 2010), this spherical parameterization method may result in significant shape distortion that affects the analysis. Our prior work (Wang et al., 2007; Wang et al., 2011; Wang et al., 2010) computed the first global conformal parameterization of lateral ventricular surfaces based on holomorphic 1-forms. However, this conformal parameterization method always introduces a singularity point (zero point, Fig. 9 (a)) in the resulting parameter domain. As a result, each ventricular surface had to be partitioned into three pieces with respect to the zero point, the superior horn, the inferior horn, and the occipital horn. These three pieces were mapped to three planar rectangles and registered across subjects separately. To model a topologically complicated ventricular surface, hyperbolic conformal geometry emerges naturally as a candidate method. Hyperbolic conformal geometry has an important property that it can induce conformal parameterizations on high-genus surfaces or surfaces with negative Euler numbers and the resulting parameterizations have no singularities (Luo et al., 2008). Motivated by recent advances in hyperbolic conformal geometry based brain imaging research (Shi et al., 2013d; Tsui et al., 2013), including our own work (Shi et al., 2012; Wang et al., 2009b; Wang et al., 2009c), here we propose to use the hyperbolic Ricci flow method to build the canonical parameter domain for ventricular surface registration. The resulting parameterizations are angle- preserving and have no singularity points. After surface registration across subjects, surface deformations are measured by the tensor-based morphometry (TBM) (Chung et al., 2008; Chung et al., 2003b; Davatzikos, 1996; Thompson et al., 2000), which quantifies local surface area expansions or shrinkages. The Ricci flow method is theoretically sound and computationally efficient (Jin et al., 2008; Wang et al., 2006; Wang et al., 2012). In addition, TBM has been used extensively to detect regional differences in surface and volume brain morphology between groups of subjects (Chung et al., 2003b; Hua et al., 2011; Leow et al., 2009; Shi et al., 2014; Shi et al., 2013a; Shi et al., 2013b; Wang et al., 2012; Wang et al., 2011; Wang et al., 2013b). We hypothesize that the hyperbolic Ricci flow together with TBM may offer a set of accurate surface statistics for ventricular morphometry and that it may boost statistical power to detect the subtle difference between MCI patients who progress to dementia from those who fail to progress. In this paper, we develop a ventricular morphometry system based on hyperbolic Ricci flow and TBM statistic and use it to study ventricular structural differences associated with baseline T1-weighted brain images from the ADNI dataset, including 71 patients who developed incident AD during the subsequent 36 months (MCI converter group) and 62 patients who did not during the same period (MCI stable group). These subjects were also selected based on the availability of fluorodeoxyglucose positron emission tomography (FDG-PET) data and cognitive assessment information. Here we set out to test whether our new system can detect subtle MCI conversion related changes and whether the new statistics are correlated with FDG-PET biomarkers and other cognitive measures. 2. SUBJECTS AND METHODS 2.1. Subjects Data used in the preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). The ADNI was launched in 2003 by the National Institute on Aging (NIA), the National Institute of Biomedical Imaging and Bioengineering (NIBIB), the Food and Drug Administration (FDA), private pharmaceutical companies and non-profit organizations, as a $60 million, 5-year public private partnership. The primary goal of ADNI has been to test whether serial magnetic resonance imaging (MRI), positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment (MCI) and early Alzheimer’s disease (AD). Determination of sensitive and specific markers of very early AD progression is intended to aid researchers and clinicians to develop new treatments and monitor their effectiveness, as well as lessen the time and cost of clinical trials. The Principal Investigator of this initiative is Michael W. Weiner, MD, VA Medical Center and University of California – San Francisco. ADNI is the result of efforts of many coinvestigators from a broad range of academic institutions and private corporations, and subjects have been recruited from over 50 sites across the U.S. and Canada. The initial goal of ADNI was to recruit 800 subjects but ADNI has been followed by ADNI-GO and ADNI-2. To date, these three protocols have recruited over 1500 adults, ages 55 to 90, to participate in the research, consisting of cognitively normal older individuals, people with early or late MCI, and people with early AD. The follow up duration of each group is specified in the protocols for ADNI-1, ADNI-2 and ADNI-GO. Subjects originally recruited for ADNI-1 and ADNI-GO had the option to be followed in ADNI-2. For up-to-date information, see www.adni-info.org. Based on the availability of both volumetric MRI and FDG-PET data, we selected 133 subjects from the MCI group in the ADNI baseline dataset, including 71 subjects (age: 74.77 ± 6.81) who developed incident AD during the subsequent 36 months, which we call the MCI converter group, and 62 subjects (age: 75.42 ± 7.83 years) who did not during the same period, which we call the MCI stable group. These subjects were chosen on the basis of having at least 36 months of longitudinal data. If a subject developed incident AD more than 36 months after baseline, it was assigned to the MCI stable group. All subjects underwent thorough clinical and cognitive assessment at the time of acquisition, including the Mini-Mental State Examination (MMSE) score (Folstein et al., 1975), Alzheimer’s disease assessment scale – Cognitive (ADAS-COG) (Rosen et al., 1984) and Auditory Verbal Learning Test (AVLT) (Rey, 1964). The demographic information of the subjects is in Table 1. MMSE at Gender (M/F) Education Age Baseline MCI Converter 45/26 15.99 ± 2.73 74.77 ± 6.81 26.83 ± 1.60 (n = 71) MCI Stable 44/18 15.87 ± 2.76 75.42 ± 7.83 27.66 ± 1.57 (n = 62) Table 1. Demographic information of studied MCI subjects in ADNI baseline dataset. 2.2. System Pipeline Overview Here we briefly overview the processing procedures in our ventricular morphometry system. Following sections are detailed explanations of each step. Figure 1 summarizes the overall sequence of steps in the system. First, from each MRI scan (a), we automatically segment lateral ventricular volumes with the multi-atlas fluid image alignment (MAFIA) method (Chou et al., 2010). The MR image overlaid with the segmented ventricle is shown in (b). A ventricular surface built with marching cube algorithm (Lorensen and Cline, 1987) is shown in (c). Lateral ventricle segmentation and surface reconstruction will be introduced in Sec. 2.3. After the topology optimization, we apply hyperbolic Ricci flow method on the ventricular surface and conformally map it to the Poincaré disk. The concepts of topology optimization and Poincaré disk model will be introduced in Sec. 2.4. Details about conformal parameterization with hyperbolic Ricci flow and embedding in Poincaré disk are in Sec. 2.5 and 2.6, respectively. On the Poincaré disk, we compute consistent geodesics and project them back to the original ventricular surface, a method called geodesic curve lifting. The results are shown in (d). Further, we convert the Poincaré model to the Klein model where the ventricular surfaces are registered by the constrained harmonic map (Zeng et al., 2010). The registration diagram is shown in (e). Geodesic curve lifting and surface registration will be detailed in Sec. 2.7. Next, we compute the TBM features and smooth them with the heat kernel method (Chung et al., 2005b) (f). TBM computation and its smoothing are in Sec. 2.8. Finally, the smoothed TBM features are applied to analyze both group difference between the two MCI groups and correlation of ventricular shape morphometry with cognitive test scores and FDG-PET index. Significance p-maps are used to visualize local shape differences or correlations (g). Correction for multiple comparisons is used to estimate the overall significance (corrected p- values). 2.3. Image Acquisition and Preprocessing High-resolution brain structural MRI scans were acquired at multiple ADNI sites using 1.5 Tesla MRI scanners manufactured by General Electric Healthcare, Siemens Medical Solutions, and Philips Medical Systems. For each subject, the T1-weighted MRI scan was collected with a sagittal 3D MP-RAGE sequence. Typical 1.5T Figure 1. A chart showing the key steps in the ventricular surface registration method. After the lateral ventricles were segmented from MRI scans and surfaces were reconstructed, we computed consistent geodesic curves on each ventricular surface to constrain the registration. Then the constrained harmonic map was used to obtain a correspondence field in the parameter domain represented by the Klein model, which also induced a surface registration in 3D. The statistic of TBM was computed on each point of the resulting matched surfaces. Finally the smoothed TBM features are applied to analyze both group difference between the two MCI groups and correlation of ventricular shape morphometry with cognitive test scores and FDG-PET index. acquisition parameters are repetition time (TR) of 2,400 ms, minimum full excitation time (TE), inversion time (TI) of 1,000 ms, flip angle of 8°, 24 cm field of view. The acquisition matrix was 192×192×166 in the x, y, and z dimensions and the voxel size was 1.25×1.25×1.2 𝑚𝑚3. In-plane, zero-filled reconstruction (i.e., sinc interpolation) generated a 256×256 matrix for a reconstructed voxel size of 0.9375×0.9375×1.2 𝑚𝑚3. The T1-weighted images from ADNI baseline dataset were automatically skull-stripped with the BrainSuite Extraction Software (Shattuck and Leahy, 2002). Then the imperfections in this automatic segmentation procedure were corrected manually. In order to adjust for global differences in brain positioning and scaling, the segmented images were normalized to the ICBM space with a 9-parameter (3 translations, 3 rotations, and 3 scales) linear transformation obtained by the Minctracc algorithm (Collins et al., 1994). After resampling into an isotropic space of 2203 voxels with the resolution 1𝑚𝑚×1𝑚𝑚×1𝑚𝑚, the registered images were then histogram-matched to equalize image intensities across subjects. Finally, the lateral ventricular volumes were extracted using the multi-atlas fluid image alignment (MAFIA) method that combines multiple fluid registrations to boost accuracy (Chou et al., 2010). Briefly, in the MAFIA method, 6 MRI scans (2 AD, 2 MCI, and 2 normal) after preprocessing were randomly chosen from the ADNI baseline dataset. The lateral ventricles were manually traced in these 6 images following the delineation protocol described in http://resource.loni.usc.edu/resources/downloads/research-protocols/segmentation/lateral-ventricle-delineation. These labeled images are called atlases and segmentation of lateral ventricles in other unlabeled images was done by fluidly registering the atlases to all other images. For details of this method, please refer to (Chou et al., 2010). After obtaining the binary segmentations of the lateral ventricles, we used a topology-preserving level set method (Han et al., 2003) to build surface models. Based on that, the marching cube algorithm (Lorensen and Cline, 1987) was applied to construct triangular surface meshes. Then, in order to reduce the noise from MR image scanning and to overcome the partial volume effects, surface smoothing was applied consistently to all surfaces. Our surface smoothing process consists of mesh simplification using “progressive meshes” (Hoppe, 1996) and mesh refinement by Loop subdivision surface (Loop, 1987). The similar procedures were frequently adopted in a number of our prior works (Colom et al., 2013; Luders et al., 2013; Monje et al., 2013; Shi et al., 2014; Shi et al., 2013a; Shi et al., 2013b; Wang et al., 2011; Wang et al., 2010) and our experience showed that the smoothed meshes are accurate approximations to the original surfaces with higher signal-to-noise ratio (SNR) (Shi et al., 2013a). 2.4. Theoretical Background This section briefly introduces the theoretical background necessary for the current work. Conformal deformation. Let 𝑆 be a surface in ℝ3 with a Riemannian metric 𝐠 induced from the Euclidean metric. Let 𝑢:𝑆 → ℝ be a scalar function defined on 𝑆. It can be verified that 𝐠̃ = 𝑒2𝑢𝐠 is also a Riemannian metric on 𝑆 and angles measured by 𝐠̃ are equal to those measured by 𝐠, i.e. the induced mapping is angle- Figure 2. Illustration of hyperbolic geometry. (a) is a pair of topological pants with three boundaries 𝛾 ,𝛾 ,𝛾 . 𝜏 ,𝜏 are 1 2 3 1 2 automatically traced paths connecting 𝛾 to 𝛾 , 𝛾 to 𝛾 , respectively. After slicing along 𝜏 ,𝜏 , the topological pants can 1 2 1 3 1 2 be conformally mapped to the hyperbolic space and isometrically embedded in the topological disk of fundamental domain, as shown in (b). (c) is an illustration of the Poincaré disk model. (d) is a saddle plane which has constant negative Gaussian curvatures with a hyperbolic triangle. preserving. Thus, 𝐠̃ is called a conformal deformation of 𝐠 and 𝑢 is called the conformal factor. Furthermore, when surface metrics change, the Gaussian curvature 𝐾 of the surface will change accordingly and become 𝐾̃ = 𝑒−2𝑢(−Δ 𝑢+𝐾), where Δ is the Laplace-Beltrami operator under the original metric 𝐠. The geodesic curvature 𝐠 𝐠 𝑘 will become 𝑘̃ = 𝑒−𝑢(𝜕 𝑢+𝑘 ), where 𝒓 is the tangent vector orthogonal to the boundary. The total 𝑔 𝑔 𝒓 𝑔 curvature of the surface is determined by its topology with the Gauss-Bonnet theorem (Do Carmo, 1976): ∫ 𝐾𝑑𝐴+∫ 𝑘 𝑑𝑠 = 2𝜋𝜒(𝑆), where 𝑑𝐴 is the surface area element, 𝜕𝑆 is the boundary of 𝑆, 𝑑𝑠 is the line 𝑆 𝜕𝑆 𝑔 element, and 𝜒(𝑆) is the Euler characteristic number of 𝑆. Uniformization theorem. Given a surface 𝑆 with Riemannian metric 𝐠, there exist an infinite number of metrics that are conformal to 𝐠. The uniformization theorem states that, among all conformal metrics, there exists a unique representative which induces constant Gaussian curvature everywhere. Moreover, the constant will be one of {+1,0,−1}. Therefore, we can embed the universal covering space of any closed surface using its uniformization metric onto one of the three canonical spaces: the unit sphere 𝕊2 for genus-0 surfaces with positive Euler characteristic numbers; the plane 𝔼2 for genus-1 surfaces with zero Euler characteristic numbers; the hyperbolic space ℍ2 for high-genus surfaces with negative Euler characteristic numbers. Accordingly, we can say that surfaces with positive Euler numbers admit spherical geometry; surfaces with zero Euler numbers admit Euclidean geometry; and surfaces with negative Euler numbers admit hyperbolic geometry. Topology optimization. Due to the concave and branching shape of the ventricular surfaces, it is difficult to find a conformal grid for the entire structure without introducing significant area distortions. Here, as in prior studies (Wang et al., 2011; Wang et al., 2010), we automatically located and introduced three cuts on each ventricular surface, with one cut on the superior horn, one cut on the inferior horn, and one cut on the occipital horn. The locations of the cuts are motivated by examining the topology of the lateral ventricles, in which several horns are joined together at the ventricular “atrium” or “trigone”. Meanwhile, we kept the locations of the cuts consistent across subjects. This operation is called as topology optimization (Wang et al., 2011; Wang et al., 2010). After being modeled in this way, each ventricular surface becomes a genus-0 surface with 3 boundaries and is homotopic to a pair of topological pants, as shown in Fig. 2 (a). Figure 4 (a) shows two different views of a ventricular surface with the three boundaries, which are denoted as 𝛾 ,𝛾 ,𝛾 . As a result, each ventricular 1 2 3 surface has the Euler characteristic number −1, which means that it admits the hyperbolic geometry. In our work, we try to compute conformal mappings from ventricular surfaces to the hyperbolic space ℍ2 and use it as the canonical parameter space to register ventricular surfaces. Poincaré disk model. As the hyperbolic space cannot be realized in ℝ3, we use the Poincaré disk model to visualize it. The Poincaré disk is the unit disk |𝑧| < 1,𝑧 = 𝑥+𝑖𝑦 in the complex plane with the metric 𝑑𝑠2 = 4𝑑𝑧𝑑𝑧̅ . The rigid motion in the Poincaré disk1 is the Möbius transformation: (1−𝑧𝑧̅)2 𝑧 → 𝑒𝑖𝜃 𝑧−𝑧0 (1) 1−𝑧̅̅0̅𝑧 A hyperbolic line (a geodesic) in the Poincaré disk is a circular arc which is perpendicular to the unit circle |𝑧| = 1. A hyperbolic circle 𝑐𝑖𝑟𝑐(𝑐,𝑟) (𝑐 is the center and 𝑟 is the radius) looks like a Euclidean circle 1In Euclidean space, rigid motions include translation and rotation. Any transformation consisting of rigid motions changes the position of an object without deforming the shape of the object. As the Poincaré disk is a representation of the hyperbolic space, the rigid motion in it is defined by Eq. (1), which is the Möbius transformation and is different from that in the Euclidean space. However, the Möbius transformation has the same properties as the Euclidean rigid motion. For example, as shown in Fig. 4 (d), the object at the center is transformed to four other different positions in the Poincaré disk with four different Möbius transformations. Each of the four pieces (shown in four colors) is a copy of the center object. They have different positions, but their shapes are the same in the hyperbolic space.