Quantifying cerebellar atrophy in multiple system atrophy of the cerebellar type (MSA-C) using three-dimensional gyrification index analysis

  • Published on
    28-Nov-2016

  • View
    225

  • Download
    10

Transcript

e stiuarsity, Taipei, Taiwan, ROCversity, TDepartment of Electrical Engineering, National Central Ue Department of Recreation Sports and Health Promotion,f Department of Computer Science and Information Enging Department of Radiology, Taipei Veterans General Hospih The Neurological Institute, Taipei Veterans General Hospi Institute of Brain Science, National Yang-Ming Universityj Department of Neurology, National Yang-Ming Universitk The Neurological Institute, Taipei Municipal Gan-Dau HoNeuroImage 61 (2012) 19Contents lists available at SciVerse ScienceDirectNeuroIm.e lMSA-C matter (3D-CBWM) to detect the atrophied cerebellar region inMSA-C patients. The 3D-GI values were in a sta-ble rangewith small variances, exhibiting no gender effect and no age-related shrinkage. Signicantly lower 3D-GI values were exhibited in both CBGM and CBWM of theMSA-C patients compared with healthy subjects, evenin the early phases of the disease. Decreases in 3D-GI values indicated the degeneration of the cerebellar foldingstructure, exactly reecting the morphological changes in cerebellum. The 3D-GI method based on CBGMresulted in superior discriminative accuracy compared with the CBGM volumetric method. Using the two-dimensional 3D-GI values, the K-means classier can evidently discriminate the MSA-C patients from healthysubjects. 2012 Elsevier Inc. All rights reserved.IntroductionMultiple system atrophy (MSA) is a form of sporadic spinocerebellardegeneration characterized by varying degrees of Parkinsonism, cerebel-lar ataxia, and autonomic dysfunction (Gilman et al., 2008; Quinn, 1989;Wenning et al., 2004). According to Gilman et al., MSA can be categorizedinto two main subtypes: MSA-C (cerebellar) and MSA-P (Parkinsonism)dysarthria, tremor, and parkinsonian or other extrapyramidal symptoms(Berciano, 1982). Many investigators have reported that the main patho-logical changes of MSA-C are the loss of neurons in the ventral portion ofthe pons, inferior olives, and cerebellar cortex. Its fundamental lesionsoccur in the arcuate, pontine, inferior olivary, pontobulbar nuclei, andthe cerebellar cortex (Pemde et al., 1995). Atrophy of the brain stemand cerebellum is an important feature on magnetic resonance (MR)(Gilman et al., 2008). In Taiwan, MSA-C is mowhile MSA-P is more common in the Unitedmost typical early symptom of this disease. Othe Corresponding authors at: Institute of Brain Science,Taipei, Taiwan, ROC. Fax: + 886 2 28273123.E-mail address: b8001071@yahoo.com.tw (P.-S. Wan1 Co-rst authors (equal contribution).1053-8119/$ see front matter 2012 Elsevier Inc. Alldoi:10.1016/j.neuroimage.2012.02.057cerebellum and brainstem atrophy more accurately. Sixteen healthy subjects and 16 MSA-C patients participat-ed in this study. We compared 3D-GI values and volumes in the cerebellum, based on T1-weighted MR images.We also compared the images of reconstructed 3D cerebellum gray matter (3D-CBGM) and cerebellum whiteAtrophyCerebellumGyrication indexa r t i c l e i n f oArticle history:Accepted 20 February 2012Available online 28 February 2012Keywords:aipei, Taiwan, ROCniversity, Chung-Li, Taiwan, ROCAsia-Pacic Institute of Creativity, Tao-Fen, Taiwan, ROCeering, National Cheng Kung University, Tainan, Taiwan, ROCtal, Taipei, Taiwan, ROCital, Taipei, Taiwan, ROC, Taipei, Taiwan, ROCy School of Medicine, Taipei, Taiwan, ROCspital, Taipei, Taiwan, ROCa b s t r a c tMultiple system atrophy of the cerebellar type (MSA-C) is a degenerative neurological disease of the centralnervous system. This study employed a method named, surface-based three-dimensional gyrication index(3D-GI) to quantify morphological changes in normal cerebellum (including brainstem) and atrophied cerebel-lum, in patients with MSA-C. We assessed whether 3D-GI can exclude gender and age differences to quantifyb Integrated Brain Research Laboratory, Departmc Brain Research Center, National Yang-Ming Unidedical Research and Education, Taipei Veterans General Hospital, Taipei, Taiwan, ROCa Department of Biomedical Imaging and Radiological Sciences, National Yang-Ming Univeent of MQuantifying cerebellar atrophy in multipl(MSA-C) using three-dimensional gyricaYu-Te Wu a,b,c,1, Kuo-Kai Shyu d,1, Chii-Wen Jao d,e, YPo-Shan Wang h,i, j,k,, Bing-Wen Soong h,j,j ourna l homepage: wwwre common than MSA-P,States. Gait ataxia is ther clinical features includeNational Yang-Ming University,g).rights reserved.ystem atrophy of the cerebellar typeon index analysisn-Lin Liao f, Tzu-Yun Wang a, Hsiu-Mei Wu g,agesev ie r .com/ locate /yn imgimages of MSA-C patients (Matsusue et al., 2008); the atrophy is aggra-vated as the disease progresses, and atrophy rates correlate negativelywith disease duration (Miyatake et al., 2010; Paviour et al., 2006).Therefore, the quantication of cerebellar atrophy is important for the di-agnosis of MSA-C and relevant studies.Neuroscientists and psychiatrists have applied several methods toassess the morphological changes of MSA-C in MR images, includingvoxel-based morphometry (VBM) (Brenneis et al., 2007; Minneropin this study. The MRI scans were conducted by the Department ofRadiology, Taipei Veterans General Hospital, Taiwan. The study wasapproved by the ethics committee of the Taipei Veterans GeneralHospital and complied with the 1964 Declaration of Helsinki. Writteninformed consent was obtained from each volunteer before the studycommenced. Diagnoses of MSA were made according to theestablished guidelines (Gilman et al., 2008). The severity of ataxiawas graded as follows: (I) walking without assistance (II) walkingwith partial assistance (III) needing assistance walking (IV) needingassistance standing and (V) bedridden (Bang et al., 2003). Theseverity of Parkinsonism was evaluated according to a modiedHoehnYahr staging system (Kim et al., 2003). The duration of illnesswas according to the patient's memory of the onset of earliest symp-toms. This was asked at their rst visit to the outpatient clinic (MRIscans were arranged at this time). The mean illness duration for thepatient group was 4.32.8 years. Demographic features and clinicaldata for the study groups are summarized in Table 1. No signicantage difference (p=0.071, t-Test) and gender difference (p=1, Chi2-Test) were revealed between the control and MSA-C groups. All pa-tients met the second consensus criteria for a clinical diagnosis ofprobable MSA (Gilman et al., 2008). According to the clinical symp-toms, patients were categorized into MSA-C.Data acquisition and image proceduresAxial magnetic resonance (MR) images of the human brain, cover-ing the entire cerebrum and cerebellum, were acquired using a 1.5-TVision Siemens scanner (Erlangen, Germany). A circularly polarizedhead coil was used, based on a T1-weighted MR sequence with the2 Y.-T. Wu et al. / NeuroImage 61 (2012) 19et al., 2007; Specht et al., 2003, 2005), volumetric analysis (VA) (Brket al., 2004; Miyatake et al., 2010) and cerebral area analysis(Horimoto et al., 2000). These methods are voxel-based measure-ment, and can be categorized as the volumetric analysis. Post-mortem and in vivo image studies have demonstrated the intrinsic bi-ological variability of sex differences and age-associated shrinkage incerebellum and brain stem volumetric analysis (Oguro et al., 1998;Raz et al., 1998, 2001; Torvik et al., 1986). The cerebellar volumes (es-pecially the right hemisphere) were larger in men, even after adjustingfor height (Raz et al., 2001). The cerebellar volume began to decrease atage 50, with progressive decrease until age 65; thereafter, the declinebecame slower and the volume stabilized at a virtual volume of90 ml (Luft et al., 1999). Male subjects showed more signicant age-related atrophy of the cerebellar hemispheres, especially the righthemisphere (Xu et al., 2000). The brainstem also showed signicantsex differences of age-associated shrinkage in different sub-regions(Oguro et al., 1998). The volumetric analysis method has limitationsin preserving the topology of cortical surface and cannot describe thecomplexity and variability of brain structural organization (Zhang etal., 2006). Accordingly, methods that can eliminate these differencesand quantify the surface complexity are preferred for assessing cere-bellar atrophy.The gyrication index (GI), proposed by Zilles et al., dened as theratio of the folded inner contours to its exposed outer contour, andhas been commonly used to measure the degree of cortical folding(Bonnici et al., 2007; Harris et al., 2004; Moorhead et al., 2006;Oyegbile et al., 2004; Zilles et al., 1988). Zilles and his colleagueshave reported that the cerebral GI values are not signicantly affectedby age, gender, body weight, body length, or brain volume differences(Zilles et al., 1988). Brain with higher degree of cortical folding yieldslarger values of GI, and these values increase proportionally to thenumber and complexity of gyri (Harris et al., 2004; Kesler et al.,2006; White et al., 2010; Zilles et al., 1997). The original GI methodrequired manual tracing of the inner and outer contours which is asubjective measure. The manual tracing procedure can be time-consuming and biased by different operators. Besides, visual inspec-tion may not be sensitive enough to trace the contours with spatiallyhigh-frequency changes. Recently, Moorhead and his colleagues havedeveloped an automated GI (A-GI) methodology to alleviate theselimitations (Moorhead et al., 2006). Differently from the manual GI,A-GI is an unbiased and objective measure and has low susceptibilityto noise (Moorhead et al., 2006). In comparison with manual tracing,A-GI substantially reduces the time costs and improves repeatability(Moorhead et al., 2006). The A-GI permits a rapid determination ofGI with unbiased in its application and unlimited in the size of test co-hort (Bonnici et al., 2007; Moorhead et al., 2006). They further dem-onstrated that the GI was sensitive to atrophy and useful forneurodevelopment measurement (Mirakhur et al., 2009). These sug-gest that GI analysis is an effective method for quantifying atrophy ofcortical folding.Previous GI studies of cortical folding were mainly focused on cere-bral and psychiatric disorders (Bonnici et al., 2007; Gaser et al., 2006;Harris et al., 2004; Zhang et al., 2009, 2010), but less in the diseaseswith atrophied cerebellums. In this study, we aim to characterize anddistinguish the cortical folding alteration of cerebellum betweenhealthy and MSA-C patients using the 3D-GI method. In addition, wemeasure the cerebellar volumes of the participants and investigatewhether there are sex difference and age effect in the 3D-GI and volu-metric measurements.Material and methodsSubjectsSixteen MSA-C patients (8 female (F)/8 male (M); mean age 567 years) and 16 healthy subjects (8 F/8M; 5110 years) participatedfollowing parameters: number of axial slices=124; slice thick-ness=1.5 mm; echo time=5.5 ms; repetition time=14.4 ms; ipangle 20; eld of view 25 cm; matrix size=256256; and in-planeresolution=1 mm1 mm.The image procedures are summarized in Fig. 1. First, the cerebel-lum (CB) image, which included the brainstem, was manuallyTable 1Demographic features and clinical data of the study groups.Demographic features Normal(16) MSA-C(16) P valueAgea(years) 5110 567 0.071Durationa (years) 4.32.8 Genderb(male/female) 8/8 8/8 1Clinical dataSeverity of ataxiacI 16(100%) 0II 0 16(100%)III 0 0IV 0 0V 0 0Modied HoehnYahr stagec0 16 01 0 02 0 16(100%)3 0 04 0 05 0 0MRI featurescNormal 16(100%) 0Cerebellar atrophy 0 16(100%)Hot cross bun sign 0 13(81%)Putaminal slits 0 5(31%)Age: age at MRI scanning. The severity of ataxia: (I) walking without assistance(II) walking with partial assistance (III) needing assistance walking (IV) needingassistance standing and (V) bedridden. MSA-C: multiple system atrophy of thecerebellar type. Modied HoehnYahr: stage of Parkinsonism.a Continuous data were expressed as meanstandard deviation (SD) and the agedifference was tested by independent t-Test.b Gender difference was tested by Chi2-Test.c Categorical data were expressed as number (%).extracted from each MR volumetric image by an experienced neurol-ogist (BW Soong and PS Wang) and re-sliced to 1 mm thicknessthrough interpolation to obtain isotropic resolution. To facilitate thismanual extraction, we built a graphical user interface using MATLAB7.0 (MathWorks, Inc., Natick, MA) which allowed us to load theimage slice by slice in the axial view, zoom in any subarea on eachslice, and draw the region of interest covering the cerebellum onthe enlarged area (Figs. 1a and f). The correlation for cerebellar vol-umes, including controls and patients, determined by two indepen-dent neurologists is 0.992 (pb0.001). The CBWM and CBGM weresegmented from the ltered CB images using the SPM5 packagefahcidjeb g 3Y.-T. Wu et al. / NeuroImage 61 (2012) 19Fig. 1. Image processes and reconstructed 3D images of CBGM and CBWM and their convexhull surfaces. (a) Manually extracted cerebellum MR image(included brainstem) of onenormal subject (b) Reconstructed 3D CBGM image (c) Reconstructed 3D CBGM convexhull surface (d) Reconstructed 3D CBWM image (e) Reconstructed 3D CBGM convex hullsurface (f) Manually extracted cerebellum MR image(included brainstem) of one MSA-Cpatient (g) Reconstructed 3D CBGM image, the red arrow indicates the anterior lobe andthe black arrow indicates the vermis region (h) Reconstructed 3D CBGM convex hull surface(i) Reconstructed 3D CBWM image, the black arrow indicates the posterior lobe and the redarrow indicates the cerebellar peduncles (e) Reconstructed 3D CBGM convex hull surface.(Wellcome Department of Imaging Neuroscience; London, UK). Thissoftware automatically segments gray matter (GM), white matter(WM), and cerebrospinal uid (CSF), and converts the segmented im-ages into binary (black-and-white) images. Second, the 3D binary im-ages of CBGM and CBWMwere reconstructed and used for volumetricand 3D-GI surface area estimation (Figs. 1b and g show CBGM;Figs. 1d and i show CBWM). Finally, the reconstructed convex hullsurfaces of 3D CBWM and CBGM were used to estimate convex hullsurface areas (Figs. 1c and h show CBGM; Figs. 1e and j show CBWM).Surface-based three dimensional gyrication indexAn alternative method for assessing brain atrophy (in addition tocerebral volume) is to compare morphological differences. To date,no golden standard has been established for quantifying shape-related difference within cerebral cortex. The original two-dimensional GI (2D-GI), proposed by Zilles et al., is the most widelyapplied method (Zilles et al., 1988). Specically, the 2D-GI is denedby the ratio of the length of complete cerebral cortex contour (includ-ing the deep sulci) to that of outer cortical contour (ignoring the deepsulci). This ratio can be formulated as:GI Pnj1Pinnerj tPnj1Pouterj twhere Pjinner is the inner perimeter for slice j, Pjouter is the outer perim-eter for slice j, t is slice thickness, and n is the number of slices in thebrain image.Although the 2D-GI method has been extensively used to deter-mine the degree of cortical folding, the cortical folding pattern mayexhibit differences in other orientations that may alter the 2D-GI indifferent slices. For example, folding along the sagittal direction maynot be apparent from a coronal or axial view. Therefore, 2D-GI cannotaccount for the inherent three-dimensional nature of the cortical sur-face (White et al., 2010).This study adopts a surface-based three-dimensional gyricationindex (3D-GI) (Rodriguez-Carranza et al., 2008), in which 3D-GIvalues are computed as the ratio of the surface area of each CBGMand CBWM partition to that of its convex hull surface. The 3D-GI for-mula used in our study is as follows:3DGI Area of 3D reconstructed CBGMCBWM surfaceArea of 3D reconstructed convex hull CBGMCBWM surface :For each reconstructed binary image object, the 3-D convex hullwas constructed by lling the local concavities in a 555 neighbor-hood iteratively. In each iteration, the border voxels (dened as back-ground voxels consisting of at least one 6-neighborhood object voxel)were rst identied; thereafter, the number of its 18-neighborhoodobject voxels in the x-, y-, and z-planes was counted, respectively.The criteria for verifying whether the current border voxel shouldbe changed from a background voxel to an object voxel were that atleast one of the three values (x-, y-, and z-planes) had to be greaterthan four, or one of them had to be equal to four with at least one8-neighbor in the same plane, with a value greater than two. According-ly, the shape of object becamemore andmore convex after certain iter-ations. The best goodness-of-t was assessed by examining whetherthe convex hull lled all the concavities.After the binarized CBGM and CBWM volumetric images and theirconvex versions were obtained, their surface areas were estimated for3D-GI calculations. The method proposed by Lindblad was adopted toestimate the surface area of the reconstructed 3D CBGM (CBWM) andtheir convex hull surfaces (Lindblad, 2003). Finally, the 3D-GI wascomputed as the ratio of the original reconstructed CBGM (CBWM)MSA-C patients manifested signicant atrophy in cerebellumFig. 1 illustrates the reconstructed 3D images of CBGM and CBWMand their convex hull surfaces, from a normal subject and a MSA-Cpatient. The anterior lobe (red arrow in Fig. 1g) of the CBGM in theMSA-C patient clearly exhibits substantial atrophy; similarly, the ver-mis (black arrow in Fig. 1g) of the MSA-C patient clearly shows wid-ening relative to that of the normal subject shown in Fig. 1b. Thecerebellar peduncle (red arrow in Fig. 1i) is slimmer in the MSA-C pa-tient, while the posterior lobe (black arrow in Fig. 1i) manifests alooser folding pattern in the CBWM of the MSA-C patient relative tothat of the normal subject (Fig. 1d).No sex difference in 3D-GI measurement for the normal groupThe CBWM volumes of the normal group showed a highly signi-cant gender difference (p=0.019; Fig. 2b). The CBGM volumes of thenormal group also exhibited a trend of gender difference (p=0.087;Fig. 2b). However, within the normal group, calculations involving3D-GI did not detect any signicant gender difference for eitherCBGM or CBWM (p=0.872 and p=0.985 respectively; Fig. 2a).No age-related shrinkage in 3D-GI measurement of the normal subjectsFigs. 3a and b depict the relationship between 3D-GI data and agefor all participants. Figs. 3c and d display the relationship betweenvolume data and age for all participants. The solid lines in Figs. 3cand d represent the regression results between volume and age ofnormal group, the black curves in Figs. 3eh denote the distributionof each data group, and the intersection of two curves is highlightedby the green dot lines.4 Y.-T. Wu et al. / NeuroImage 61 (2012) 19surface area derived by their reconstructed convex hull surface area.The volumes of 3D CBWM and CBGM were also estimated by thetotal pixel count within the targets, and displayed in cm3 units.Statistical analysesFirstly, the homogeneity of variances of CBGM and CBWM 3D-GI(and volumes) between compared groups were veried using Mau-chly's test of sphericity (Mauchly, 1940). The one-way ANOVA analy-sis was used to assess the gender difference within the data of thenormal group, for 3D-GI values and CBGM and CBWM volumes. Theage correlations between 3D-GI values and volumes within the nor-mal group were also examined. To determine if there is a signicantdifference between the normal and patient groups for 3D-GI valuesof CBWM (and CBGM), and if there is a gender effect between gendergroups, this study adopted the two-way ANOVA analysis in which therst factors were CBWM 3D-GI or CBGM 3D-GI, and the second factorwas gender. The two levels for the rst factors were MSA-C group andnormal group, and male and female for the second factor. Similarly,the between-group statistics for CBWM volumes and CBGM volumes,and the gender effect were also analyzed using the two-way ANOVA.The rst factors were CBWM volumes or CBGM volumes, with twolevels being MSA-C group and normal group, and the second factorwas gender, with two levels being male and female. In addition, wecalculated the correlation coefcients between the 3D-GI values andvolumetric values in normal subjects and MSA-C patients. The corre-lation between 3D-GI values (and volumes) and duration of MSA-Cpatients were also examined. The signicance level for all of thetests was set at pb0.05. All the statistics were calculated using theStatistics Toolbox in MATLAB 7.0.Discriminative accuracy assessment of 3D-GI and volumetric measuresIn addition to statistical analyses, the discriminative accuracy of3D-GI (and volume) was assessed by setting the cut-off values ofeach measure between normal and MSA-C groups. We further usethe CBGM and CBWM 3D-GI (and volumes) as paired parametersand feed them into an unsupervised 3D-GI based (and volumebased) classier, namely, the K-means classier (Bishop, 1995). Theclustering results of the K-means classier (3D-GI based or volumebased) may indicate which measure is superior in discriminatingMSA-C patients from normal subjects.ResultsFirstly, each variance of CBWM and CBGM 3D-GI (and volumes)between normal and MSA-C group was tenable in the assumption ofsphericity (p=0.8866 for CBGM 3D-GI, p=0.1157 for CBWM 3D-GI, p=0.2347 for CBGM volume and p=0.2198 for CBWM volume),which allowed us to use the ANOVA analysis appropriately.No trend of age difference between normal and MSA-C groupsAlthough our result indicated that p-value of age difference waslarger than 0.05 (p=0.071, t-Test), we further evaluated whetherthere was a trend towards age difference in this study. We removedsubjects younger than 40 years old from the normal group and there-by the remaining 13 subjects (6 males and 7 females) were formed asthe revised normal group. The age difference between the revisednormal and MSA-C groups was insignicant (p=0.389, t-Test). Thestatistical result showed both the 3D-GI and volumetric values stillremained signicant difference between the revised normal andMSA-C groups (p=6.45*1012 for CBGM 3D-GI, p=5.68*107 forCBWM 3D-GI, p=0.0105 for CBGM volume and p=5.77*105 forCBWM volume). The result suggested that there was no trend of agedifference in this study.CBGM CBWM11.522.533D-GIMaleFemalea p=0.872p=0.985CBGM CBWM406080100120140Volume(cm3 )MaleFemaleb p=0.087p=0.019Fig. 2. Gender difference of CBGM and CBWM data within normal subjects. M/F:male/female, CBWM: cerebellum white matter, CBGM: cerebellum gray matter, CB:cerebellum (a) Comparison of CBGM and CBWM GI values between genders withinnormal subjects. (b) Comparison of CBGM and CBWM volumes between genderswithin normal subjects.30 40 50 60 701.822.22.42.62.8Age at scanning(years) CBGM 3D-GINormal CBGMMSA-C CBGMa0 0.1 0.2 0.3 0.41.822.22.42.62.8ProbabilityCBGM 3D-GI Normal CBGM MSA-C CBGMcut-off value=2.3e30 40 50 60 7011.11.21.31.41.51.6Age at scanning(years) CBWM 3D-GINormal CBWMMSA-C CBWMb0 0.1 0.2 0.3 0.411.11.21.31.41.51.6ProbabilityCBWM 3D-GI Normal CBWM MSA-C CBWMfcut-off value=1.3530 40 50 60 706080100120140160Age at scanning(years)CBGM Volume(cm3 )Normal CBGMMSA-C CBGMc ( r=-0.4967, p=0.0503 )0 0.1 0.2 0.3 0.46080100120140160ProbabilityCBGM Volume(cm3 ) Normal CBGM MSA-C CBGMcut-off value=110g30 40 50 60 7010203040506070Age at scanning(years)CBWM Volume(cm3 )Normal CBWMMSA-C CBWM( r=-0.5467, p=0.0284 )d0 0.1 0.2 0.3 0.410203040506070ProbabilityCBWM Volume(cm3 ) Normal CBWM MSA-C CBWMcut-off value=41hFig. 3. Scatter plots of CBGM or CBWM 3D-GI values (and volumes) versus age and distribution curves of CBGM or CBWM 3D-GI values (and volumes) for each group. Figs. 3adindicate the scatter plots of all 3D-GI values and volume values versus age, solid lines represent the regression results between the normal group volumetric data and age. Figs. 3ehindicate the distribution of all 3D-GI values and volume values for each group, the black curves represent the distribution of 3D-GI values (or volumes) of each group and the greendot lines indicated the intersected values of the intersected curves.5Y.-T. Wu et al. / NeuroImage 61 (2012) 19The regression lines in Figs. 3c and d indicate statistically signi-cant age-related shrinkages in the volumetric measurements of bothCBGM and CBWM for the normal subjects (r=0.4967, p=0.0503and r=0.5467, p=0.0284, respectively). In contrast, no statisticalcorrelations were found between age and GI values (for eitherCBGM or CBWM) of the normal subjects.Signicant smaller standard deviation and stable range of 3D-GI valuesfor the normal subjectsFor normal subjects, the ratio of standard deviation (SD) to meanGI was 3.9% for CBGM and 4.2% for CBWM, respectively. These valuesare signicantly smaller than the ratios of SD to mean volume, whichare 9.1% for CBGM and 12.6% for CBWM (Table 2). There was no sta-tistically signicant correlation between 3D-GI values and volume innormal subjects (r=0.2015, p=0.4542 for CBGM; r=0.1954,p=0.4682 for CBWM).Signicantly decreased 3D-GI and volumetric values for the MSA-C groupThe estimated 3D-GI values (meanstandard deviation) fornormal subjects were 2.51690.1058 for CBGM and 1.4222CBWM 3D-GI and CBWM volume have the same ratio (87.5%),suggesting that these two measures have comparable of discrimination.Figs. 5(a) and (b) display the clustering results for CBGM andCBWM 3D-GIs, and that for CBGM and CBWM volumes, respectively.In Fig. 5(a), normal and MSA-C groups can be accurately classiedinto two clusters in which all the normal subjects are at the upper-right region (blue squares) and the MSA-C subjects are at thelower-left region (red circles). In Fig. 5(b), the K-means methodmisclassies three normal subjects (red circle with black star inside)as MSA-C patients, and one MSA-C patient (blue square with greenstar inside) as normal subject.DiscussionTo our knowledge, this is the rst study to apply the surface-based3D-GI method to investigate changes in cortical folding complexity inthe normal and atrophied cerebellum. The estimated 3D-GI values ofthe normal group were relatively stable, with a small variance, andshowed no signicant sex difference or age-related shrinkage.Because the GI is a relative measure, a sex difference in this parameterwould be expected only if there were accompanying sex-dependentdifferences in the basic intracortical structure (Zilles et al., 1988).The applied surface-based method can accurately extract the6 Y.-T. Wu et al. / NeuroImage 61 (2012) 190.0549 for CBWM (Table 2). The 3D-GI values (meanstandard de-viation) of the MSA-C patients were 2.02970.1189 for CBGM and1.27220.0685 for CBWM (Table 3). The two-way ANOVA resultsare summarized in Table 4 and Fig. 4. The 3D-GI values for the MSA-C group were signicantly smaller than those of the normal groupfor both CBGM and CBWM (p=1.18*1012 and p=3.29*107 re-spectively; Fig. 4a). The CBGM and CBWM volumes of MSA-C subjectswere also signicantly smaller than those of the normal subjects.(p=0.0014 for CBGM and p=2.65*106 for CBWM; Fig. 4b).Signicantly decreased 3D-GI values in early duration of MSA-C patientsWe had chosen seven MSA-C patients whose duration was lessthan 3 years (please refer to Table 3) as the early duration MSA-Cgroup and preformed comparison of 3D-GIs and volumes with normalgroup. The CBGM 3D-GI, CBWM 3D-GI and CBWM volume all showedsignicant decrease, while the CBGM volume showed no signicantdecrease (p=1.61*109 for CBGM 3D-GI, p=1.81*105 forCBWM 3D-GI, p=0.003 for CBWM volume and p=0.246 for CBGMvolume). The correlation analysis between the duration of MSA-CTable 23D-GI value and volume of CBWM and CBGM for each normal subject.Subjects Sex Age(years) CBWMvol.CBGMvol.CBWM3D-GICBGM3D-GICB vol.1 M 34 58.1 140.5 1.3405 2.5623 198.72 M 36 55.5 133.5 1.4279 2.3685 189.13 M 44 52.6 124.7 1.4079 2.2948 177.34 M 45 44.9 115.9 1.3728 2.5548 162.55 M 55 48.5 129.4 1.4925 2.6481 177.96 M 55 46.4 116.1 1.3437 2.4344 162.57 M 57 60.1 129.5 1.4856 2.6676 189.78 M 65 46.6 111.5 1.5123 2.6417 158.19 F 39 52.9 132.3 1.5019 2.5685 185.310 F 47 48.8 111.5 1.3877 2.4633 160.311 F 48 40.1 100.8 1.3842 2.4203 140.912 F 50 45.8 106.8 1.4257 2.4222 152.613 F 54 43.1 117.8 1.3900 2.5554 161.014 F 55 42.4 125.0 1.3972 2.5667 167.515 F 57 49.8 122.7 1.4658 2.5350 172.616 F 70 39.9 110.0 1.4187 2.5668 149.9Average 51 48.5 120.5 1.4222 2.5169 169.12S D 10 6.1 10.97 0.0549 0.1058 16.14Percent of SD 12.6% 9.1% 3.9% 4.2% 9.55%Note: M/F: male/female, CBWM: cerebellum white matter, CBGM: cerebellum graymatter, CB: cerebellum, Vol.: volume (cm3), SD: standard deviation.patients and 3D-GI (and volume) showed only that the CBGM volumehad signicant correlation with duration (p=0.0079), while theCBGM 3D-GI, CBWM 3D-GI and CBWM volume exhibited no signi-cant correlation with the duration (p=0.2253 for CBGM 3D-GI,p=0.5430 for CBWM 3D-GI and p=0.1922 for CBWM volume).Superior discriminative accuracy of 3D-GI measureIn Figs. 3eh, the green dashed lines indicate the intersected valueof two distribution curves. We dened the intersected value as thecut-off values for discriminating the MSA-C patients and normal.The discriminative accuracy of the cut-off values was evaluated bythe ratio of the number of normal subjects' 3D-GI (or volume) largerthan the cut-off value to the total number of normal subjects. Amongthese four measurements, the CBGM 3D-GI produces the highest ratio(93.75%), indicating that it is a superior discriminative measurement.On the other hand, the CBGM volume results in the lowest ratio(81.25%), and thereby an inferior discriminative measurement. TheTable 33D-GI value and volume of CBWM and CBGM for each MSA-C patient.Patient Sex Age(years)Duration(years)CBWMvol.CBGMvol.CBWM3D-GICBGM3D-GICBvol.1 M 49 6 19.4 68.7 1.2602 2.1819 88.12 M 51 3 34.8 114.4 1.2687 2.1694 149.33 M 57 7 26.9 78.2 1.2976 1.9721 105.14 M 58 5 27.1 104.3 1.2016 2.1177 131.55 M 60 3 35.2 102.9 1.2654 2.1301 138.16 M 63 5 33.6 105.6 1.1898 1.9992 139.37 M 67 1 58.1 139.4 1.3965 1.9582 197.58 M 68 8 45.4 88.8 1.3869 1.9728 134.29 F 47 5 24.7 91.9 1.2691 2.0263 116.710 F 48 2 33.2 112.1 1.2950 1.8462 145.411 F 49 1 25.2 113.1 1.2669 1.9189 138.312 F 50 9 26.7 108.5 1.1226 2.2758 135.313 F 53 1 34.1 113.7 1.2235 2.0992 147.814 F 54 4 34.6 110.1 1.3102 1.9622 144.715 F 61 1 36.8 105.0 1.2796 1.9185 141.816 F 68 8 28.2 90.7 1.3214 1.9265 118.9Average 56 4.3 32.8 103.0 1.2722 2.0297 135.8SD 7 2.8 9.2 16.6 0.0685 0.1189 23.5Note: M/F: male/female, Duration: duration in years, CB: cerebellum, CBWM:cerebellum white matter, CBGM: cerebellum gray matter, Vol.: volume (cm3), CB:cerebellum, SD: standard deviation.topology of cortical surface and continuous structural alteration.The regression lines in Figs. 3c and d show signicant age-relatedshrinkage of the cerebellum, according to volumetric measurement.This age-related shrinkage associated with volumetric measurement,and gender effect, limits the assessment of neuronal degeneration.Due to age-related shrinkage or large volumetric variance betweenindividuals, it is usually difcult to dene age-related shrinkage ordisease atrophy for a single subject based on volumetric measure-ment alone. This is particularly true for older subjects because age-related shrinkage may be more signicant for them than for youngersubjects. By contrast, the applied 3D-GI method is a ratio measure-ment of cortical complexity and is independent of age-relatedshrinkage.The results showed that for healthy elderly subjects, age-relatedshrinkage may reduce the size of cerebellum but not the cortical com-plexity. Zilles et al. indicated that no signicant decrease of GI wasfound in the cerebral cortex during the aging process up to 91 years(Zilles et al., 1997); hence the CBGM and CBWM 3D-GI values ofhealthy subjects all fall within a narrow range, with small variation.neurological abnormality. These characteristics make the 3D-GI valuea good descriptor of cortical folding. Furthermore, the stable rangeand exclusion of age-related shrinkage in 3D-GI values effectuate de-ning the normal 3D-GI range of CBGM and CBWM reliably.Our nding of gray matter loss in the upper vermis and anteriorlobe, and white matter loss in the peduncles and posterior lobe, isin line with the previous MSA study (Specht et al., 2003). Lesions inthe posterior vermis or anterior lobe of the cerebellum may affectstance and gait in humans (Thach and Bastian, 2004). Gyral atteningand reduced sulcal depth are symptoms of neuronal degenerationTable 4Two-way ANOVA for the analysis of between-group volumes, between-group 3D-GI, and the gender effect.Normal MSA-C P valueMale(n=8) Female(n=8) Male(n=8) Female(n=8) df error-df F 3D-GI GenderCBGM 2.5210.139 2.5120.066 2.0630.096 1.9970.136 1 28 146.85 1.18*1012 0.359CBWM 1.4230.068 1.4220.045 1.2830.076 1.2610.064 1 28 44.15 3.29*107 0.618Male(n=8) Female(n=8) Male(n=8) Female(n=8) df error-df F Volume GenderCBGM 125.19.97 115.810.40 100.2922.03 105.639.28 1 28 12.57 0.0014 0.687CBWM 51.585.83 44.744.42 35.0612.04 30.444.75 1 28 34.37 2.65*106 0.038The rst factors are CBGM volume (cm3), CBWM volume (cm3), CBGM 3D-GI and CBWM 3D-GI, respectively. Two levels for the rst factors are normal group and MSA-C group. Thesecond factor is gender with two levels which are male and female.1.1 1.2 1.3 1.4 1.5 1.61.822.22.42.62.8CBWM 3D-GICBGM 3D-GINormal SubjectMSA-C Subjectcluster 1cluster 2a7Y.-T. Wu et al. / NeuroImage 61 (2012) 19Alteration in the 3D-GI value may indicate a change in cortical com-plexity, impairment or atrophy in the brain, and the degree of specicCBGM CBWM11.522.533D-GINormalMSA-Cap=1.18*10-12p=3.29*10-7140 NormalMSA-Cp=0.0014bCBGM CBWM20406080100120Volume(cm3 )p=2.65*10-6Fig. 4. Two-way ANOVA results for the normal group and MSA-C group. (a) Comparison ofCBGM and CBWM 3D-GI values between normal group and MSA-C group. (b) Comparisonof CBGM and CBWM volumes between normal group and MSA-C group.80100120140160CBGM Volume(cm3 )Normal SubjectMSA-C Subjectcluster 1cluster 2b10 20 30 40 50 60 7060CBWM Volume(cm3)Fig. 5. Classication results using the K-means classier method. The data of normal andMSA-C patients are denoted by the black crosses and the green stars, respectively. Theresultant clusters 1 and 2 are denoted by the red circles and blue squares, respectively.(a) 3D-GI based K-means classier, all the MSA-C patients were classied in the cluster 1and all the normal subjects were classied in the cluster 2 (b) Volume based K-meansclassier, three normal subjects were misclassied in the cluster 1(red circle with blackcross inside) and one MSA-C patient was misclassied in the cluster 2(blue square withgreen star inside). The distribution of MSA-C and normal subjects is clearly separated inFig. 5(a) and accurately classied into 2 clusters. On the contrary, there is overlaps andmisclassication of MSA-C and normal subjects by using volume based classier methodin Fig. 5(b).Ogata, K., Kawai, M., 2010. Brain volume analyses and somatosensory evoked8 Y.-T. Wu et al. / NeuroImage 61 (2012) 19disorders. Even a tiny volumetric decrease leads to a decrease in theinner contour surface, and hence lowers the 3D-GI values. Althoughboth the 3D-GI values and volumetric values of MSA-C patientsexhibit signicant differences from the normal group, the 3D CBGMand CBWM of MSA-C patients show fewer folding patterns and asmaller volumetric size, hence smaller 3D-GI values and volumetricvalues are anticipated.A prominent nding of this study is that MSA-C patients demon-strate signicantly more hypogyrication in both the CBGM andCBWM surface of the cerebellum than normal subjects do, even inthe early phases of the disease. Moreover, the analysis results ofcut-off values and K-means classier veried that the 3D-GI can bea superior discriminator between normal and MSA-C patients. Incomparison with the CBGM and CBWM volumes, the much smallerp-values of CBGM and CBWM 3D-GI, especially in the early duration,indicate that the 3D-GI method may provide a promising way to dis-tinguish an atrophied cerebellum. Overall the 3D-GI outperforms thevolumetric method in discriminating normal subjects from MSA-Cpatients.The CBWM and CBGM 3D-GI values showed signicant decreasein early duration, but revealed insignicant correlation with duration.The superior sensitivity and age irrelevance of 3D-GI measure wouldexplain such a discrepancy. In the early duration, the 3D-GI measureis more sensitive than the volumetric measure in detecting the tinyalteration of cerebellar atrophy. As the duration and age both in-crease, the cerebellar atrophy due to disease severity and age-related shrinkage becomes worse, which allow the volumetric mea-sure to detect the alteration and therefore the volumetric measureshows a signicant correlation with duration. On the other hand,the MSA-C patients' cerebellar atrophied rate can be subject depen-dent and may not directly relate to the disease duration since somepatients may have atrophy faster than others under the same dura-tion. Accordingly, the CBWM and CBGM 3D-GI values presented insig-nicant correlation with duration that was not surprising.The fractal dimension (FD) is another popular method for quanti-fying the sulcal/gyral folding pattern of brain cortex (Esterban et al.,2007, 2009; King et al., 2010; Wu et al., 2009, 2010; Zhang et al.,2006). The FD analysis, proposed by Mandelbrot, can condense allthe structural details of an irregular object into a single numericvalue and serve as a quantitative measure of morphological complex-ity (Mandelbrot, 1977, 1982). The uses of FD are diversied and havebeen applied on clinical images analysis, such as tumor detection(Iftekharuddin et al., 2000; Pereira et al., 2000; Zook andIftekharuddin, 2005) and respiration system analysis (Peng et al.,2002). In comparison with the FD, GI is limited to the application ofmeasuring the cortical folding based on its derivation from corticalconvolution. King et al. reported that the GI is positively correlatedwith FD and both are suitable for quantifying folding complexity(King et al., 2010). However, precise image segmentation and deter-mination of suitable range of box sizes should be carried out toavoid degrading the accuracy and sensitivity in the FD method.In conclusion, the 3D-GI analysis can avoid gender effect, age-related shrinkage, and produce results with smaller variances. TheMSA-C patients exhibit signicantly lower 3D-GI values both inCBGM and CBWM. The 3D-GI method is more suitable for the assess-ment of morphological change because of its superior sensitivity anddiscrimination.AcknowledgmentsThis study was supported by research grants from Taipei VeteransGeneral Hospital (V99C1-052), National Science Council (NSC 99-2221-E-075-003, NSC 099-2811-E-010-002), NSC support for theCenter for Dynamical Biomarkers and Translational Medicine, NationalCentral University, Taiwan (NSC 100-2911-I-008-001) and BrainResearch Center, National Yang-Ming University and a grant frompotentials in multiple system atrophy. J. Neurol. 257, 419425.Moorhead, T.W., Harris, J.M., Staneld, A.C., Job, D.E., Best, J.J.K., Johnstone, E.C., Lawrie,S.M., 2006. Automated computation of the yrication index in prefrontal lobes:method and comparison with manual implementation. NeuroImage 31, 15601566.Oguro, H., Okada, K., Yamaguchi, S., Kobayashi, S., 1998. Sex differences in morphol-ogy of the brain stem and cerebellum with normal ageing. Neuroradiology 40,Ministry of Education, Aim for the Top University Plan. We thankMr. Valerio Puggioni (Wallace Academic Editing) for English improvingand editing of this study. Authors thank anonymous reviews for theirinsightful comments and suggestions.ReferencesBang, O.Y., Huh, K., Lee, P.H., Kim, H.J., 2003. Clinical and neuroradiological features ofpatients with spinocerebellar ataxias from Korean kindreds. Arch. Neurol. 60,15661574.Berciano, J., 1982. Olivopontocerebellar Atrophy. A review of 117 cases. J. Neurol. Sci.53, 253272.Bishop, C.M., 1995. Neural Networks for Pattern Recognition. Oxford University,Oxford, U.K.Bonnici, H.M., Moorhead, T.W.J., Staneld, A.C., Harris, J.M., Owens, D.G., Johnstone,E.C., Lawrie, S.M., 2007. Pre-frontal lobe gyrication index in schizophrenia, mentalretardation and comorbid groups: an automated study. NeuroImage 35, 648654.Brenneis, C., Egger, K., Scherer, C., Seppi, K., Schocke, M., Poewe, W., Wenning, G.K.,2007. Progression of brain atrophy in multiple system atrophy: a longitudinalVBM study. J. Neurol. 254, 191196.Brk, K., Globas, C., Wahl, T., Bhring, U., Diety, K., Yhlke, C., Luft, A., Schuly, J.B., Voigt,K., Dichgans, J., 2004. MRI-based volumetric differentiation of sporadic cerebellarataxia. Brain 127, 175181.Esteban, F.J., Sepulcre, J., de Mendizbal, N.V., Goi, J., Navas, J., de Miras, J.R., Bejarano,B., Masdeu, J.C., Villoslada, P., 2007. Fractal dimension and white matter changes inmultiple sclerosis. NeuroImage 36, 543549.Esteban, F.J., Sepulcre, J., de Miras, J.R., Navas, J., de Mendizabal, N.V., Goni, J., Quesada,J.M., Bejarano, B., Villoslada, P., 2009. Fractal dimension analysis of grey matter inmultiple sclerosis. J. Neurol. Sci. 282, 6777.Gaser, C., Luders, E., Thompson, P.M., Lee, A.D., Dutton, R.A., Geaga, J.A., Hayashi, K.M., Bellugi,U., Galaburda, A.M., Korenberg, J.R., Mills, D.L., Toga, A.W., Reiss, A.L., 2006. Increasedlocal gyrication mapped in Williams syndrome. NeuroImage 33, 4654.Gilman, S., Wenning, G.K., Low, P.A., Brooks, D.J., Mathias, C.J., Trojanowski, J.Q., Wood,N.W., Colosimo, C., Drr, A., Fowler, C.J., Kaufmann, H., Klockgether, T., Lees, A.,Poewe, W., Quinn, N., Revesz, T., Robertson, D., Sandroni, P., Seppi, K., Vidailhet,M., 2008. Second consensus statement on the diagnosis of multiple systematrophy. Neurology 71, 670676.Harris, J.M., Yates, S., Miller, P., Best, J.J.K., Johnstone, E.C., Lawrie, S.M., 2004. Gyricationin rst-episode schizophrenia: a morphometric study. Biol. Psychiatry 55, 141147.Horimoto, Y., Aiba, I., Yasuda, T., Ohkawa, Y., Katayama, T., Yokokawa, Y., Goto, A., Ito,Y., 2000. Cerebral atrophy in multiple system atrophy by MRI. J. Neurol. Sci. 173,109112.Iftekharuddin, K.M., Jia, W., Marsh, R., 2000. A fractal analysis approach to identicationof tumor in brain MR images. Eng. Med. Biol. Soc., Proc. of the 22nd Annu. Intl.Conference of the IEEE, 4, pp. 30643066.Kesler, S.R., Vohr, B., Schneider, K.C., Katz, K.H., Makuch, R.W., Reiss, A.L., Ment, L.R.,2006. Increased temporal lobe gyrication in preterm children. Neuropsychologia44, 445453.Kim, S.E., Choi, J.Y., Choe, Y.S., Choi, Y., Lee, W.Y., 2003. Serotonin transporters inthe midbrain of Parkinson's disease patients: a study with 123I--CIT SPECT.J. Nucl. Med. 44, 870876.King, R.D., Brown, B., Hwang, M., Jeon, T., George, A.T., 2010. Fractal dimension analysisof the cortical ribbon in mild Alzheimer's disease. NeuroImage 53, 471479.Lindblad, J., 2003. Surface area estimation of digitized planes using weighted localcongurations. Lect. Notes Comput. Sci. 2886, 348357.Luft, A.R., Skalej, M., Schulz, J.B., Welte, D., Kolb, R., Brk, K., Kolckgether, T., Voigt, K.,1999. Patterns of age-related shrinkage in cerebellum and brainstem observed invivo using three-dimensional MRI volumetry. Cereb. Cortex 9, 712721.Mandelbrot, B.B., 1977. Fractals: Form, Chance, andDimension.W.H. Freeman, San Francisco.Mandelbrot, B.B., 1982. The Fractal Geometry of Nature. W.H. Freeman, SanFrancisco.Matsusue, E., Fujii, S., Kanasaki, Y., Sugihara, S., Miyata, H., Ohama, E., Ogawa, T., 2008.Putaminal lesion in multiple system atrophy: postmortem MR-pathologicalcorrelations. Neuroradiology 50, 559567.Mauchly, J.W., 1940. Signicance test for sphericity of a normal n-variate distribution.Ann. Math. Stat. 11, 204209.Minnerop, M., Specht, K., Ruhlmann, J., Schimke, N., Abele, M., Weyer, A., Wllner, U.,Klockgether, T., 2007. Voxel-based morphometry and voxel-based relaxometry inmultiple system atrophya comparison between clinical subtypes and correlationswith clinical parameters. NeuroImage 36, 10861095.Mirakhur, A., Moorhead, T.W.J., Staneld, A.C., Mckirdy, J., Sussmann, E.D., Hall, J.,Lawrie, S.M., Johnstone, E.C., Mclntosh, A.M., 2009. Changes in gyrication over4 years in bipolar disorder and their association with brain-derived neurotrophicfactor valine66 methionine variant. Biol. Psychiatry 66, 293297.Miyatake, S., Mochizuki, H., Naka, T., Ugawa, Y., Tanabe, H., Kuzume, D., Suzuki, M.,788792.Oyegbile, T., Hansen, R., Magnotta, V.O., Leary, D., Bell, B., Seidenberg, M., Hermann,B.P., 2004. Quantitativemeasurement of cortical surface features in localization-related temporal lobe epilepsy. Neuropsychology 18, 729737.Paviour, D.C., Price, S.L., Jahanshahi, M., Lees, A.J., Fox, N.C., 2006. Longitudinal MRI inprogressive supranuclear palsy and multiple system atrophy: rates and regionsof atrophy. Brain 129, 10401049.Pemde, H.K., Bakhshi, S., Kalra, V., 1995. Olivopontocerebellar atrophy: a case report.Brain Dev. 17, 130132.Peng, C.K., Mietus, J.E., Liu, Y., Lee, C., Jeffrey, M., Hausdorff, J.M., Stanley, H.E., Goldberger,A.L., Lipsitz, L.A., 2002. Quantifying fractal dynamics of human respiration: age andgender effects. Ann. Biomed. Eng. 30, 683692.Pereira, D., Zambrano, C., Martin-Landrove, M., 2000. Evaluation of malignancy intumors of the central nervous system using fractal dimension. Eng. Med. Biol.Soc. Proc. of the 22nd Annu. Intl. Conference of the IEEE, 3, pp. 17751778.Quinn, N., 1989. Multiple system atrophythe nature of the beast. J. Neurol. Neurosurg.Psychiatry 52, 7889.Raz, N., Dupuis, J.H., Briggs, S.D., McGavran, C., Acker, J.D., 1998. Differential effects ofage and sex on the cerebellar hemispheres and the vermis: a prospective MRstudy. AJNR Am. J. Neuroradiol. 19, 6571.Raz, N., Gunning-Dixon, F., Head, D., Williamson, A., Acker, J.D., 2001. Age and sexdifferences in the cerebellum and the ventral pons: a prospective MR study ofhealthy adults. AJNR Am. J. Neuroradiol. 22, 11611167.Rodriguez-Carranza, C.E., Mukherjee, P., Vigneron, D., Barkovich, J., Studholme, C.,2008. A framework for in vivo quantication of regional brain folding in prematureneonates. NeuroImage 41, 462478.Specht, K., Minnerop, M., Abele, M., Reul, J., Wllner, U., Klockgether, T., 2003. In vivovoxel-based morphometry in multiple system atrophy of the cerebellar type.Arch. Neurol. 60, 14311435.Specht, K., Minnerop, M., Wller-Hbenthal, J., Thomas Klockgether, T., 2005. Voxel-based analysis of multiple-system atrophy of cerebellar type: complementaryresults by combining voxel-based morphometry and voxel-based relaxometry.NeuroImage 25, 287293.Thach, W.T., Bastian, A.J., 2004. Role of the cerebellum in the control and adaptation ofgait in health and disease. Prog. Brain Res. 143, 353366.Torvik, A., Torp, S., Lindboe, C.F., 1986. Atrophy of the cerebellar vermis in ageing: amorphometric and histological study. J. Neurol. Sci. 76, 283294.Wenning, G.K., Colosimo, C., Geser, F., Werner Poewe, W., 2004. Multiple systematrophy. Lancet Neurol. 3, 93103.White, T., Su, S., Schmidt, M., Kao, C.Y., Sapiro, G., 2010. The development of gyricationin childhood and adolescence. Brain Cogn. 72, 3645.Wu, Y.T., Shyu, K.K., Chen, T.R., Guo, W.Y., 2009. Using three-dimensional fractaldimension to analyze the complexity of fetal cortical surface from magneticresonance images. Nonlinear Dyn. 58, 745752.Wu, Y.T., Shyu, K.K., Jao, C.W., Wang, Z.Y., Soong, B.W., Wu, H.M., Wang, P.S., 2010. Fractaldimension analysis for quantifying cerebellar morphological change of multiplesystem atrophy of the cerebellar type (MSA-C). NeuroImage 49, 539551.Xu, J., Kobayashi, S., Yamaguchi, S., Iijima, K.I., Okada, K., Yamashita, K., 2000. Gender effectson age-related changes in brain structure. AJNR Am. J. Neuroradiol. 21, 112118.Zhang, L., Dean, D., Liu, J.Z., Sahgal, V., Wang, X., Yue, G.H., 2006. Quantifying degener-ation of white matter in normal aging using fractal dimension. Neurobiol. Aging 28,15431555.Zhang, Y., Yu, C., Zhou, Y., Li, K., Li, C., Jiang, T., 2009. Decreased gyrication in majordepressive disorder. Neuroreport 20, 378380.Zhang, Y., Zhou, Y., Yu, C., Lin, L., Li, C., Jiang, T., 2010. Reduced cortical folding in mentalretardation. Am. J. Neuroradiol. 31, 10631067.Zilles, K., Armstrong, E., Schleicher, A., Kretschmann, H.J., 1988. The human pattern ofgyrication in the cerebral cortex. Anat. Embryol. 179, 173179.Zilles, K., Schleicher, A., Langemann, C., Amunts, K., Morosan, P., Palomero-Gallagher,N., Schormann, T., Mohlberg, H., Burgel, U., Steinmetz, H., Schlaug, G., Roland,P.E., 1997. Quantitative analysis of Sulci in the human cerebral cortex: development,regional heterogeneity, gender difference, asymmetry, intersubject variability andcortical architecture. Hum. Brain Mapp. 5, 218221.Zook, J.M., Iftekharuddin, K.M., 2005. Statistical analysis of fractal-based brain tumordetection algorithms. Magn. Reson. Imaging 23, 671678.9Y.-T. Wu et al. / NeuroImage 61 (2012) 19Quantifying cerebellar atrophy in multiple system atrophy of the cerebellar type (MSA-C) using three-dimensional gyrification index analysisIntroductionMaterial and methodsSubjectsData acquisition and image proceduresSurface-based three dimensional gyrification indexStatistical analysesDiscriminative accuracy assessment of 3D-GI and volumetric measuresResultsNo trend of age difference between normal and MSA-C groupsMSA-C patients manifested significant atrophy in cerebellumNo sex difference in 3D-GI measurement for the normal groupNo age-related shrinkage in 3D-GI measurement of the normal subjectsSignificant smaller standard deviation and stable range of 3D-GI values for the normal subjectsSignificantly decreased 3D-GI and volumetric values for the MSA-C groupSignificantly decreased 3D-GI values in early duration of MSA-C patientsSuperior discriminative accuracy of 3D-GI measureDiscussionAcknowledgmentsReferences