CIVET-Chimp: An automated pipeline for MRI-based cortical surface extraction in chimpanzees Authors: Claude Lepage, Erika Nolan, Trisanna Sprung-Much, Gabriel A. Devenyi, Michael Petrides, Alan C. Evans Published: 2026 ABSTRACT The MNI CIVET pipeline for automated extraction of cortical surfaces and evaluation of cortical thickness from in vivo human magnetic resonance imaging has recently been extended to process macaque brains and, following this work, has since been broadened to process chimpanzee (Pan troglodytes) brains. The pipeline uses the NCBR (National Chimpanzee Brain Resource) 3 T dataset, the largest high-resolution chimpanzee magnetic resonance imaging collection, and processing is performed in a derived standardized MNI-like NCBR reference space. An average population surface, constructed from the cortical surfaces of the individual NCBR subjects, serves as the basis for surface-based registration for group comparisons and ROI-based analyses, with integrated Davi130 and Brainnetome parcellations. A surface-based transformation is provided for mapping common human average surfaces (MNI152 and fsavg) to the MNI-NCBR chimpanzee average surface to facilitate inter-species studies. The modifications needed to adapt CIVET from the macaque to the chimpanzee brain are detailed and cortical thickness maps obtained for the NCBR subjects are compared to other published maps. The open usage of CIVET-Macaque and CIVET-Chimp is expected to promote collaborative efforts in data collection and processing, automated analyses, and data sharing, thereby contributing to advances in the field of non-human primate brain imaging. Introduction With its close phylogenetic proximity to humans, chimpanzees (Pan troglodytes) represent a highly valuable model to examine evolutionary similarities and differences in brain organization (Rilling, 2014, Friedrich et al., 2021). During the past few decades, there has been growing interest in exploring the structural characteristics of the chimpanzee brain, both independently and in comparison to the human brain, notably with respect to cortical morphometry (Donahue et al., 2018, Hopkins and Avants, 2013, Hopkins et al., 2017b, Vickery et al., 2024, Xiang et al., 2020), cytoarchitecture (Gallardo et al., 2023, Palomero-Gallagher and Zilles, 2019, Schenker et al., 2010, Semendeferi et al., 2001), white matter connectivity (Bryant et al., 2020, Chauvel et al., 2024), and sulcal morphology (Amiez et al., 2023a, Amiez et al., 2019, Hathaway et al., 2024, Hopkins et al., 2021, Hopkins et al., 2022, Keller et al., 2012, Miller et al., 2020, Willbrand et al., 2023). These neuroanatomical studies have also offered crucial information on the development and specialization of cognitive functions between species. For example, Amiez et al. (2023b) recently identified a unique anatomical feature of the human frontal operculum that likely contributed to the emergence of speech in humans, distinguishing them from non-human primates. The study of the chimpanzee brain thus provides an essential comparative framework that contributes to advancing our understanding of the evolution of the organization of the human cerebral cortex, including possible structural-functional relationships. Surface-based analysis of the cerebral cortex using MRI intrinsically respects the two-dimensional topology of the cortex, providing an optimal and advantageous approach to examine its structural properties (Dale et al., 1999, Fischl et al., 1999). To extend such an analysis in the context of non-human primates (NHP), multiple computational pipelines have been made openly available on the PRIM-atE Resource Exchange (PRIME-RE) (Messinger et al., 2021) for the processing of NHP neuroimaging data. However, these specialized pipelines have primarily been targeted towards the processing of the macaque, without any customization for other non-human primates, such as the chimpanzee and the marmoset. Furthermore, given that no pipeline has been fully tailored specifically for the processing of chimpanzee MRI data, tissue segmentation and surface generation must often be performed using segregated human or NHP software tools. As a result, a number of researchers have used FSL (Jenkinson et al., 2012), CAT12, SPM12, NHP-HCP, or some custom algorithms for preprocessing, followed by an adaptation of the NHP or human FreeSurfer pipelines (FreeSurferNHP, Human-FreeSurfer) for tissue segmentation and surface reconstruction (Donahue et al., 2018, Autio et al., 2020, Gallardo et al., 2023, Hathaway et al., 2024, Hopkins et al., 2017b, Miller et al., 2020, Willbrand et al., 2023). The pipeline is completed by surface-based registration using, for instance, Multimodal Surface Matching (MSM) (Robinson et al., 2014) for corticometric analyses. Another human brain MRI analysis suite that has been used to reconstruct chimpanzee cortical surfaces, after various adjustments and workarounds, is BrainVISA (Mangin et al., 2004), which can also be used to extract sulci and perform corticometric analyses (Hopkins et al., 2017a, Hopkins et al., 2022). Alternatively, computational pipelines Precon-all and deep learning based FastSurfer (Henschel et al., 2020) are also promising tools, but need further validation as they have not yet been applied directly to chimpanzee data in a large capacity. In other words, given that human and existing NHP pipelines are not directly suitable for chimpanzee neuroimaging, this has led many research groups to develop their own in-house procedures consisting substantially of pre- and post-processing adjustments. Collectively, these challenges underscore the need for an open-access pipeline specifically tailored to chimpanzee MRI data and aligned with human imaging standards. To meet the increasing interest in chimpanzee neuroimaging research, and to overcome the limitations of the existing methodologies mentioned above, we have developed a novel pipeline for the automated MRI-based cortical surface extraction of chimpanzee (Pan troglodytes) brains. This pipeline, called CIVET-Chimp, is an extension of the Montréal Neurological Institute (MNI) CIVET (Lepage et al., 2017) and CIVET-Macaque (Lepage et al., 2021) pipelines to process human and macaque MRI data, respectively. Processing within the CIVET-Chimp pipeline is performed in high-resolution species-specific volumetric and surface spaces based on datasets from the National Chimpanzee Brain Resource (NCBR) (https://www.chimpanzeebrain.org) (Devenyi et al., 2020). The CIVET pipeline suite thus serves as a powerful tool for enabling comparative studies of human, macaque, and chimpanzee MRI data within a fully integrated framework. The CIVET-Chimp pipeline presented here is the first fully integrated and automated structural magnetic resonance imaging pipeline specifically tailored for the cortical surface extraction of chimpanzee brains based on species-specific models within a standardized chimpanzee space. It performs registration, segmentation, and surface reconstructions of T1-weighted anatomical scans, generates high-resolution white, pial, and mid cortical surfaces, and provides surface morphometrics such as cortical thickness, surface curvature, and areas. Of particular importance, the pipeline was used to create an average surface template derived from the cortical surfaces of the complete NCBR 3 T MRI dataset. This average surface can be used for surface-based registration at the group level and for multimodal analyses. The open usage of the CIVET-Chimp pipeline will be the catalyst needed to propel collaborative efforts in the fields of non-human primate research and translational neuroimaging. Material and methods NCBR dataset The NCBR offers the largest and richest openly available dataset of chimpanzee (Pan troglodytes) structural MRI scans and includes multimodal data such as diffusion tensor imaging, as well as postmortem specimens. This is particularly relevant given the limited accessibility to previously collected scans and new chimpanzee neuroimaging data since the 2015 implementation of the US Fish and Wildlife Service and the National Institute of Health (NIH) regulations governing research using chimpanzees. High-resolution T1-weighted magnetic resonance image in vivo scans at 1.5 T and 3 T were made available through NCBR (supported by the NIH grant NS092988). The cohort of animals scanned at 3 T (Yerkes colony) were housed at the Yerkes National Primate Research Center (YNPRC), while the animals scanned at 1.5 T (Bastrop colony) were housed at the National Center for Chimpanzee Care (NCCC) (see Table 1 for a summary). The 3 T subjects (N = 77, 51 females, 26 males, ranging from 6 to 54 years) were acquired using a 3.0 T scanner (Siemens Trio, Siemens Medical Solutions USA, Inc.) at the YNPRC. The T1-weighted images were collected using a three-dimensional gradient echo sequence with minor variations in sequences to accommodate the different subjects (pulse repetition between 2300 and 2600 ms, echo time from 3.0 to 4.4 ms, voxel resolution from 0.60 to 0.90 mm isotropic, number of signals averaged from 1 to 3, and a matrix size of either 320 x 320 x 192 or 256 x 256 x 128). The 1.5 T subjects (N = 138, 78 females, 60 males, ranging 9 to 51 years) were scanned using a 1.5 T Phillips (The Netherlands) or GE (Milwaukee, WI) machine at the NCCC. T1-weighted images were collected in the transverse plane using a gradient echo protocol (pulse repetition 23 ms, echo time 78 ms, number of signals averaged 2, 0.625 mm x 0.625 x 1.2 mm resolution, and a 256 x 256 matrix size). The confirmation of the left and right orientation was based on the verification of the location of a vitamin placed during the initial scan. The subjects have been anonymized for the purpose of publication. No new magnetic resonance imaging data was acquired for the current study. Data collection was approved by the Institutional Animal Care and Use Committees at the YNPRC and the NCCC, part of the University of Texas MD Anderson Cancer Center, and followed the Institute of Medicine guidelines on the ethical treatment of chimpanzees in research.Table 1Description of NCBR datasets for the Bastrop and Yerkes colonies.Table 1ColonyField strengthNSexAge range (yr)Bastrop1.5 T13878F/60M28.37±10.12Yerkes3 T7751F/26M24.77±11.35Total215129F/86M27.08±10.69 Table 1Description of NCBR datasets for the Bastrop and Yerkes colonies.Table 1ColonyField strengthNSexAge range (yr)Bastrop1.5 T13878F/60M28.37±10.12Yerkes3 T7751F/26M24.77±11.35Total215129F/86M27.08±10.69 Description of NCBR datasets for the Bastrop and Yerkes colonies. Average volumetric template The unique subjects used for the creation of the average volumetric template were selected from the 3 T scans on the basis of the contrast-to-noise ratio (CNR) (Magnotta et al., 2006), disregarding the 1.5-T scans (see Fig. 1), resulting in a volume averaged from 75 subjects (50 females, 25 males, 24.46 ± 11.34 years) (Devenyi et al., 2020). DICOM files from the NCBR dataset were converted to MINC (Vincent et al., 2016) and reoriented when necessary to the standard RAS coordinates. In the case of repeated scans (within the same session), the scans were averaged, per subject, to improve signal-to-noise ratio (SNR). Each scan was then registered to the MNI_ICBM_NLIN_SYM_09c symmetric human model (Fonov et al., 2009) using a full affine transform, which was de-scaled and de-sheared to produce a robust rigid-body transform which was used to bring the scan into an MNI-like standardized orientation while maintaining voxel dimensions. Finally, the resampled scans were processed through a custom iterative bias field/masking / classification pipeline iterativeN4_multispectral.sh (Tustison et al., 2010), using the prior MNI model, to produce the pre-processed scans of the first round. These first round scans were skull stripped, averaged and the resulting average was resampled at 0.4 mm isotropic voxel dimensions to use as the starting target for the application of antsMultivariateTemplateConstruction2.sh (Avants et al., 2011). The input brains were cut from the skull and subjected to rigid, then affine, then nonlinear template construction with a modified image pyramid (see Table 2 with 4 rounds of template convergence at each level, using the volgenmodel method of Janke and Ullmann (2015)). Masks and tissue classifications from the first round pre-processing were resampled, averaged, and smoothed in the first-round average space to produce chimpanzee priors. The average volume and priors were aligned with the AC-PC line of the MNI space and used for a second round of pre-processing to produce the final classifications, tissue priors, and masks. The resulting volumetric template is symmetric, readily amenable for the detection and analysis of brain asymmetries. It has been resampled at an isotropic voxel size of 0.4 mm and lies in a standardized orientation similar to MNI (Devenyi et al., 2020).Fig. 1Contrast-to-noise ratio for the 1.5T and 3T scans of the full NCBR dataset.Fig. 1Table 2Image pyramid parameters for non-linear template construction of the NCBR average volume using the volgenmodel method from Janke and Ullmann (2015).Table 2Cycle#IterationsShrinksSmoothing sigmas (mm)(σ=0.4/2ln2=0.3397)120251012σ220251011σ320251010σ4202599σ5202588σ6202577σ6202576σ7202565σ8202554σ9202543σ1090033σ1130022σ121001σ1310010 Contrast-to-noise ratio for the 1.5T and 3T scans of the full NCBR dataset. Table 2Image pyramid parameters for non-linear template construction of the NCBR average volume using the volgenmodel method from Janke and Ullmann (2015).Table 2Cycle#IterationsShrinksSmoothing sigmas (mm)(σ=0.4/2ln2=0.3397)120251012σ220251011σ320251010σ4202599σ5202588σ6202577σ6202576σ7202565σ8202554σ9202543σ1090033σ1130022σ121001σ1310010 Image pyramid parameters for non-linear template construction of the NCBR average volume using the volgenmodel method from Janke and Ullmann (2015). CIVET-chimp implementation The implementation of CIVET-Chimp parallels the implementation of CIVET-Macaque (Lepage et al., 2021), with species-specific models, masks, tissue priors, and atlases defined over the NCBR template used for stereotaxic space registration, regional masking, tissue classification, surface-based registration, and surface parcellation for ROI-based analyses (see Fig. 2). CIVET-Chimp requires on input a mandatory T1-weighted in vivo scan (in sphinx position), with optional T2-weighted and proton density scans, which can provide a more robust multispectral tissue classification. Mirroring the work for the implementation of CIVET-Macaque, a length scale of 0.60 relative to the human brain size is introduced for the chimpanzee (0.40 for the macaque, 1.0 for the human). This length scale is used for adjusting length parameters, for example, the FWHM during blurring, to account for the size and morphology of the brain of the given species.Fig. 2Preparation of NCBR model: (A) brain mask (red); (B) regional masks for ventricles (dark blue), subcortical region (red), brainstem (green), cerebellum (not visible in this plane), hippocampus (cyan), and amygdala (not visible in this plane); (C) subset of the points used to define priors for the three tissue classes: CSF (red), GM (yellow), WM (blue).Fig. 2 Preparation of NCBR model: (A) brain mask (red); (B) regional masks for ventricles (dark blue), subcortical region (red), brainstem (green), cerebellum (not visible in this plane), hippocampus (cyan), and amygdala (not visible in this plane); (C) subset of the points used to define priors for the three tissue classes: CSF (red), GM (yellow), WM (blue). Processing includes affine registration in the NCBR standardized space (Collins et al., 1994), non-uniformity corrections using N3 (Sled et al., 1998), brain masking using a NHP-tailored FSL bet implementation (Smith, 2002), tissue classification (Zijdenbos et al., 1998, Tohka et al., 2004), extraction of white and pial cortical surfaces (Kim et al., 2005), evaluation of cortical thickness (Lerch and Evans, 2005), and surface-based registration to the MNI-NCBR average surface (Lyttelton et al., 2007, Boucher et al., 2009) (see Section 2.4). Table 3 summarizes the various implementation differences for the three species represented in the CIVET pipeline. The raw white surfaces are extracted in the standardized NCBR volumetric space using a marching-cubes algorithm, then immediately resampled at a standard resolution of 40,962 vertices per hemisphere via interpolation to the inflated sphere (refined icosahedral topology). The white surface, first obtained from the discrete tissue classification, is adjusted at the maximum T1w intensity gradient defining the gray-white border. The pial surface is obtained by expanding the white surface to the pial boundary. Vertices between the white and pial surfaces match and the vertex numbering between left- and right-oriented surfaces is mirrored, thus enabling direct asymmetry analyses. Surface-based registered thicknesses can then be resampled on this surface template for group analyses. Regional analyses are made possible using the Davi130 parcellation, which has been mapped from the Juna.Chimp brain (Vickery et al., 2020) to the MNI-NCBR average surface, or using the Chimpanzee Brainnetome Atlas (CBA) parcellation (Wang et al., 2025), which has been mapped from the Yerkes29 v1.2 surface to the MNI-NCBR average surface (see Section 2.5). Quality control images are provided to evaluate the performance of the pipeline. Brain masking for macaques and chimpanzees is particularly challenging due to large differences in the thickness of the scalp and muscles around the head across males, females, and juveniles. Since brain masking is performed in standardized space, the mask of the NCBR model is first warped onto the subject’s brain to restrict the region where intensity thresholds are to be evaluated. With these optimal tissue thresholds, the brain can be successfully extracted using a customized version of FSL bet (Smith, 2002) with processing parameters scaled by the brain size of the species (see Table 4).Table 3Differences in processing steps across the CIVET-Human, CIVET-Chimp and CIVET-Macaque configurations. The length scale is defined as a linear factor relative to the human brain size. The FWHM on surfaces is scaled to account for the gyral patterns of the given species.Table 3Processing StepHumanChimpMacaqueStereotaxic templateMNI152-ICBM, ADNI,NCBRNMTColin27Stereotaxic voxel sampling0.50 or 1.0 mm0.40 or 0.80 mm0.25 or 0.50 mmVolumetric registrationlength scale 1.0length scale 0.6length scale 0.4 (fwhm, simplex, sampling)Neck cropping (headheight)175 mm110 mm50 mmMasking (search distance)length scale 1.0length scale 0.6length scale 0.4Non-uniformity correctionsspline distance 125 (3 T)spline distance 50spline distance 50Classificationpriors on human templatepriors on NCBRpriors on NMTSegmented maskson human templateon NCBRon NMT (brainstem, ventricles, cerebellum, etc.)Surface extractionlength scale 1.0length scale 0.6length scale 0.4 (fwhm, gradient, dilation/erosion)Surface resolution (vertices)40,962 or 163,84240,962 or 163,84240,962 or 163,842Cortical thickness methodtlink, tlaplace, tfstlink, tlaplace, tfstlink, tlaplace, tfsCortical thickness blurring30 mm fwhm12 mm fwhm8 mm fwhmSurface-basedMNI152MNI-NCBRNMT registration templateSurface-based registrationlength scale 1.0length scale 0.6length scale 0.4 (fwhm)Surface parcellationAAL, DKT, lobesDavi130, CBA, lobesCHARM, D99, lobes Table 3Differences in processing steps across the CIVET-Human, CIVET-Chimp and CIVET-Macaque configurations. The length scale is defined as a linear factor relative to the human brain size. The FWHM on surfaces is scaled to account for the gyral patterns of the given species.Table 3Processing StepHumanChimpMacaqueStereotaxic templateMNI152-ICBM, ADNI,NCBRNMTColin27Stereotaxic voxel sampling0.50 or 1.0 mm0.40 or 0.80 mm0.25 or 0.50 mmVolumetric registrationlength scale 1.0length scale 0.6length scale 0.4 (fwhm, simplex, sampling)Neck cropping (headheight)175 mm110 mm50 mmMasking (search distance)length scale 1.0length scale 0.6length scale 0.4Non-uniformity correctionsspline distance 125 (3 T)spline distance 50spline distance 50Classificationpriors on human templatepriors on NCBRpriors on NMTSegmented maskson human templateon NCBRon NMT (brainstem, ventricles, cerebellum, etc.)Surface extractionlength scale 1.0length scale 0.6length scale 0.4 (fwhm, gradient, dilation/erosion)Surface resolution (vertices)40,962 or 163,84240,962 or 163,84240,962 or 163,842Cortical thickness methodtlink, tlaplace, tfstlink, tlaplace, tfstlink, tlaplace, tfsCortical thickness blurring30 mm fwhm12 mm fwhm8 mm fwhmSurface-basedMNI152MNI-NCBRNMT registration templateSurface-based registrationlength scale 1.0length scale 0.6length scale 0.4 (fwhm)Surface parcellationAAL, DKT, lobesDavi130, CBA, lobesCHARM, D99, lobes Differences in processing steps across the CIVET-Human, CIVET-Chimp and CIVET-Macaque configurations. The length scale is defined as a linear factor relative to the human brain size. The FWHM on surfaces is scaled to account for the gyral patterns of the given species. The CIVET pipeline suite operates using the MINC format for volume (and MNI obj for surfaces) (Vincent et al., 2016), a highly efficient file format for medical imaging files featuring internal file compression (using HDF-5 (The HDF Group, 2016)) and supporting metadata. Additionally, the pipeline supports input volumes in the popular NIfTI format and, optionally, conveniently provides output in the NIfTI and GIfTI formats. The pipeline is currently openly accessible through the CBRAIN platform (Sherif et al., 2014).Table 4FSL bet processing parameters for the human, chimpanzee, and macaque brain extraction, customized for species brain size. Smaller search distances and radii of curvature are specified for smaller brain sizes.Table 4ParameterHumanChimpMacaqueIterations200020002000Cycles101010Cost Search7.03.02.5Cost Upper Search1.00.50.4Brain Threshold0.50.50.5Max Radius10.07.56.0Min Radius3.332.52.0 Table 4FSL bet processing parameters for the human, chimpanzee, and macaque brain extraction, customized for species brain size. Smaller search distances and radii of curvature are specified for smaller brain sizes.Table 4ParameterHumanChimpMacaqueIterations200020002000Cycles101010Cost Search7.03.02.5Cost Upper Search1.00.50.4Brain Threshold0.50.50.5Max Radius10.07.56.0Min Radius3.332.52.0 FSL bet processing parameters for the human, chimpanzee, and macaque brain extraction, customized for species brain size. Smaller search distances and radii of curvature are specified for smaller brain sizes. Average surface template The symmetric average MNI-NCBR surface (40,962 vertices per hemisphere) was created from the left/right surfaces extracted from 74 T1-weighted MRI scans of 73 subjects from the 3 T NCBR dataset (one repeated scan for 0065 was included and subjects 0074, 0075, 0076, and 0077 were excluded after quality control). The 1.5 T scans did not provide suitable tissue contrast for reliable extraction of cortical surfaces and were therefore not included in the generation of the surface template. The final sample consisted of 48 females and 25 males (mean age of 24.75 ± 10.88 years, repeated subject 0065 is male). The average population symmetric surface enables direct surface-based registration of individual chimpanzee surfaces to a true chimpanzee template and, consequently, allows chimpanzee-specific surface parcellations to be used for ROI-based analyses. Additionally, cross-species mappings can be obtained across the human, chimpanzee, and macaque, via surface-based registration for comparative studies. The raw white surfaces were first extracted for the NCBR 3 T subjects using CIVET-Chimp. Since the surfaces are in a common volumetric standardized space and have the same number of vertices (icosahedral mesh with 6-level subdivision), inflation to sphere yields similar vertex-wise projections onto the sphere and a direct average across all subjects in a coordinate-wise manner provides an immediate meaningful average surface at the initial iteration of registration. A symmetric average surface is obtained by mirroring the surfaces of the individual subjects before averaging. The individual white surfaces are then registered to this first average, based on cortical folding patterns, resampled, and averaged again. Equal-size triangles are imposed to provide uniform sampling across the surface. At the next iteration, the surfaces are registered, surface-based, to the average surface at the previous iteration. This registration-averaging process (Lyttelton et al., 2007) is repeated for 10 iterations until convergence is reached (the first 5 iterations are without gray-white intensity gradient correction, the last 5 iterations are with the gradient correction). The white average surface is used during surface extraction to define a mask of the medial plane cut of the corpus callosum and brainstem over which the gray-white matter T1w intensity gradient calibration is disabled. Using this final average white surface in CIVET-Chimp, the complete set of raw surfaces, white and pial, are re-extracted, from which the mid-cortical surfaces are obtained. The generation of the average mid-cortical surface follows the same iterative procedure as its average white surface counterpart, with the specification that the mid-cortical average surface is co-registered to the average white surface. This ensures that the average white and mid-cortical surfaces are in the same vertex-wise space. The mid-cortical average surface is used for the classic surface-based registration to a common template. High-resolution versions are available at 163842 vertices per hemisphere. The extraction of the white surfaces for the individual NCBR subjects required manual annotations of the tissue classification to correct, mainly, for lack of intensity contrast and the presence of high-intensity voxels representing blood vessels in the T1w images. This ensured that the true topology of the white surface was preserved, without bridges in the surface across the sulci. Although the CIVET pipeline is meant to be fully automatic, it is possible for the user to intervene at some stages to correct for anomalies in tissue classification due to artifacts in the T1w input scans. The convergence history for the creation of the symmetric average MNI-NCBR surface is depicted in Fig. 3 for the white and mid-cortical surfaces. The average surface of the gray-white matter boundary is shown in Fig. 4. The variability in shape (mean vertex displacement) in the N = 74 scans is highest in the parieto-occipital area and in the medial and ventral occipital lobe. The average mid-cortical surface exhibits these same patterns as the average white matter surface. Such variability can be explained by the inter-subject differences in morphology, with additional contributions from processing due to the limited gray-white tissue contrast in the T1w images. Fig. 3Convergence of the average symmetric white matter (left panel) and mid-cortical (middle panel) surfaces for the NCBR cohort (N = 74 scans). The residual represents the vertex-wise mean displacement in the average surface from one iteration to the next and should tend to zero as the vertices are no longer moving. Max represents the largest vertex-wise displacement from one iteration to the next. The sudden peak at the 5th iteration of the white surface creation is due to the activation of the gray-white T1w intensity gradient. Deviation represents the mean vertex-wise displacement for the N = 74 scans (148 surfaces) and is a measurement of inter-subject shape variability. The rightmost panel shows the histogram of the area of the triangles on the final average white matter and mid-cortical surfaces, indicating a nearly uniform point sampling over the surface.Fig. 3Fig. 4Views of the average white symmetric surface (top two rows) and color map of mean vertex displacement for the N = 74 scans of the NCBR dataset (bottom two rows).Fig. 4 Convergence of the average symmetric white matter (left panel) and mid-cortical (middle panel) surfaces for the NCBR cohort (N = 74 scans). The residual represents the vertex-wise mean displacement in the average surface from one iteration to the next and should tend to zero as the vertices are no longer moving. Max represents the largest vertex-wise displacement from one iteration to the next. The sudden peak at the 5th iteration of the white surface creation is due to the activation of the gray-white T1w intensity gradient. Deviation represents the mean vertex-wise displacement for the N = 74 scans (148 surfaces) and is a measurement of inter-subject shape variability. The rightmost panel shows the histogram of the area of the triangles on the final average white matter and mid-cortical surfaces, indicating a nearly uniform point sampling over the surface. Views of the average white symmetric surface (top two rows) and color map of mean vertex displacement for the N = 74 scans of the NCBR dataset (bottom two rows). Surface parcellations Cortical parcellations and atlases are a valuable tool for ROI-based analyses. Such parcellations are to be defined on the average population surface and mapped to individual subjects via surface-based registration. The present work includes two such cortical parcellations of the chimpanzee brain, namely the Davi130 atlas (Vickery et al., 2020) and the Chimpanzee Brainnetome Atlas (CBA) (Wang et al., 2025), although other publicly available parcellations of the chimpanzee brain are available, such as the von Economo and Bonin &Bailey cortical atlas (EBB-38), a cytoarchitectonic atlas that describes 38 homologous cortical areas (per hemisphere) between humans and chimpanzees (van den Heuvel et al., 2019). For their usage in CIVET-Chimp, these parcellations must first be mapped to the MNI-NCBR average surface. The Davi130 atlas is defined over the Juna.Chimp volume and consists of up to 130 parcels representing the cortex and subcortical structures. For mapping the atlas onto the MNI-NCBR average surface, the NCBR volume was first registered non-linearly to the Juna.Chimp volume. Then, this volumetric NCBR-Juna transformation was applied to the surface-based registered resampled mid-cortical surfaces (left and right) of the 74 NCBR scans to warp them into Juna space. These warped resampled surfaces were intersected with the Davi130 atlas in volumetric space to produce vertex-wise labels on the surfaces, which were then averaged across the 74 surfaces for each hemisphere. The most probable labels were kept on a per vertex basis. Because surface-based registered resampled surfaces were used, the labels obtained are automatically defined on the MNI-NCBR average surface. Finally, a few manual interventions were necessary to clean up the final parcellation due to noise resulting from erroneous segmentation of the Davi130 volumetric atlas. Although the MNI-NCBR average surface is symmetric, the resulting parcellation on this average surface is not symmetric, since the Davi130 volumetric atlas is itself not symmetric. The resulting parcellation, with 47 cortical regions per hemisphere, is shown in Fig. 5. The nomenclature of the 47 regions is described in Figure 2 of Vickery et al. (2020). The CBA asymmetric parcellation consists of 200 cortical regions defined over the Yerkes29 v1.2 reference average surface (29 females from the 3 T NCBR dataset). Direct surface-based registration based on folding patterns (using sulcal depth) was used to map the parcellation from the Yerkes29 average surface to the MNI-NCBR average surface. The resulting (asymmetric) parcellation, with 100 cortical regions per hemisphere, is shown in Fig. 6. The nomenclature of the regions is listed in Supplementary Table 2 of Wang et al. (2025).Fig. 5The Davi130 atlas overlaid on the Juna.Chimp volume (axial and coronal views presented) and on the MNI-NCBR average surface (lateral and medial views presented). The parcellation includes 47 cortical regions per hemisphere.Fig. 5 The Davi130 atlas overlaid on the Juna.Chimp volume (axial and coronal views presented) and on the MNI-NCBR average surface (lateral and medial views presented). The parcellation includes 47 cortical regions per hemisphere. Fig. 6The Chimpanzee Brainnetome Atlas (CBA) overlaid on the Yerkes29 v1.2 volume (axial and coronal views presented) and on the MNI-NCBR average surface (lateral and medial views presented). The parcellation includes 100 cortical regions per hemisphere. The color map for the CBA has been reproduced from the original publication of Wang et al. (2025).Fig. 6 The Chimpanzee Brainnetome Atlas (CBA) overlaid on the Yerkes29 v1.2 volume (axial and coronal views presented) and on the MNI-NCBR average surface (lateral and medial views presented). The parcellation includes 100 cortical regions per hemisphere. The color map for the CBA has been reproduced from the original publication of Wang et al. (2025). Surface-based registration to the human brain The growing interest in studying NHP to gain insight into the evolution of the human brain raises the inevitable need to map the cortex from one species to the latter. Already, the various implementations of NHP pipelines based on the FreeSurfer environment set of tools make direct use of human-based models, namely the fsaverage model, for surface-based registration (Hopkins et al., 2017b) using mris_register based on surface curvature. Recently, a voxel-based approach has been employed for defining diffeomorphic transformations across the human, chimpanzee, and macaque brains (Amano et al., 2025). Foreseeably, following work on the alignment of the cortex of the macaque brain to the human brain (Xu et al., 2020) (Yerkes19 to fs_LR), formal mappings are sought to map the MNI-NCBR average population surface to standardized human templates (FreeSurfer’s fsaverage and CIVET’s MNI152). The intuitive direct approach would be to map the average surfaces of each species to one another – human to chimpanzee – using surface-based registration based on cortical folding patterns (curvature, sulcal depth), but difficulties soon arise due to the differences in morphological shape between these human and chimpanzee average brain surfaces. In this work, an alternate approach was adopted, to help circumvent the aforementioned shortcomings, in which each cortical surface of the N = 74 NCBR scans was registered to the human template, in occurrence the CIVET MNI152 average surface. The length scale parameters for surface-based registration across the chimpanzee and human brains were self-adjusted based on the total surface area of each brain (e.g. the FWHM for surface blurring on each surface is scaled based on the square root of the ratio of surface area of the two surfaces). This approach proved to be robust as the morphology of individual chimpanzee brains sufficiently resembles that of the human average surface in terms of major gyral and sulcal landmarks, justifying the use of cortical folding patterns (curvature, sulcal depth) for registration. The 74 surface-based registered chimpanzee surfaces were resampled in the MNI152 surface space, then averaged altogether to form an average chimpanzee surface in the MNI152 space. This latter surface, chimpanzee shaped, was ultimately registered to the MNI-NCBR average surface, establishing a global transformation from the MNI-NCBR surface space to the MNI152 surface space. The large number of subjects (N = 74) made this approach more robust against outliers due to local mis-registration across surfaces originating from different species. At last, the MNI-NCBR-to-MNI152 transformation was concatenated to the MNI152-to-fsaverage transformation (human-to-human) to obtain the mapping from MNI-NCBR surface space to the popular fsaverage. Transformations are presented in the customary spherical representation. Results Cortical surfaces were obtained from the 74 scans of the 73 subjects from the NCBR 3T Yerkes cohort (one repeated scan). The 1.5T NCBR scans were not processed due to very low CNR (see Fig. 1). The surfaces were produced with 40,962 vertices per hemisphere. Fig. 7 shows a graphical overview of the CIVET-Chimp stages and typical results for 4 NCBR subjects (2 males, 2 females, at different ages). The tfs metric (thickness like FreeSurfer) was used for the calculation of cortical thickness, isotropically blurred at a FWHM of 16 mm. Emphasis should be directed at the low gray-white contrast in the T1w scans, with possible degradation of the cortical thickness measurements. The average cortical thickness map and its coefficient of variation (COV=standard deviation/mean) on the 74 NCBR Yerkes scans are shown in Fig. 8. The cortical thickness patterns show lower values in the postcentral gyrus and occipital lobe, as expected, but the high COV values in the precentral gyrus, in particular, are likely more a reflection of the poor gray-white T1w intensity contrast than the variability between subjects itself in that region. Such limitations in the surface extraction process due to scan quality (low CNR) must be taken into account when interpreting such local cortical thickness maps for this dataset.Fig. 7CIVET-Chimp results for 4 NCBR subjects. Within each set of image, for a single subject, the top row, from left to right: coronal slice of the T1w image, brain mask, tissue classification (CSF in blue, cortical GM in light green, portions of non-cortical GM in white, WM and other non-cortical GM in yellow), and cortical surfaces overlaid on the T1w image (WM surface in red, pial surface in green); bottom row, from left to right: white and pial surfaces; cortical thickness (range 0.5 mm blue to 3 mm red, blurred 16 mm fwhm) and the CBA parcellation viewed on the pial surface (100 ROIs).Fig. 7Fig. 8Mean cortical thickness (top) and coefficient of variation (COV) of cortical thickness (bottom) for the scans of the NCBR Yerkes dataset (N = 74), viewed on the MNI-NCBR average white cortical surface. Cortical thickness was computed from the tfs metric and smoothed over the surface using a FWHM of 16 mm.Fig. 8 CIVET-Chimp results for 4 NCBR subjects. Within each set of image, for a single subject, the top row, from left to right: coronal slice of the T1w image, brain mask, tissue classification (CSF in blue, cortical GM in light green, portions of non-cortical GM in white, WM and other non-cortical GM in yellow), and cortical surfaces overlaid on the T1w image (WM surface in red, pial surface in green); bottom row, from left to right: white and pial surfaces; cortical thickness (range 0.5 mm blue to 3 mm red, blurred 16 mm fwhm) and the CBA parcellation viewed on the pial surface (100 ROIs). Mean cortical thickness (top) and coefficient of variation (COV) of cortical thickness (bottom) for the scans of the NCBR Yerkes dataset (N = 74), viewed on the MNI-NCBR average white cortical surface. Cortical thickness was computed from the tfs metric and smoothed over the surface using a FWHM of 16 mm. In supplementary analyses, the mean cortical thickness for each subject was plotted as a function of age in Fig. 9, revealing the expected trend in atrophy of the cortex with age in the chimpanzee, in good agreement with Demirci et al. (2023) (see Fig. 11, for N = 54 of the 3T NCBR subjects). A comparison of the gyrification index (GI), based on the pial surface per hemisphere, across macaque, chimpanzee, and human populations is given in Fig. 10, where it can be observed, as expected, that surface complexity increases within the hierarchy of primates. For the purpose of illustration of the human-to-chimpanzee alignment, the CBA atlas was transposed to the widely used fsaverage human cortical surface (see Fig. 11). The qualitatively excellent alignment of the ROIs, with matching structural boundaries between the chimpanzee and human brains, suggests that the mapping can be used with confidence for comparative studies. Conversely, it would be a straightforward exercise to map cortical parcellations defined on the human brain onto the chimpanzee brain.Fig. 9Mean cortical thickness vs age for the NCBR Yerkes dataset (N = 74).Fig. 9Fig. 10Comparison of gyrification index (GI) on the pial surface per hemisphere across the macaque (NMT, N = 29 (Seidlitz et al., 2018), chimpanzee (NCBR, N = 74), and human (MNI152-extended, N = 251 (Kötter et al., 2001)) datasets. Mean and standard deviation of GI increase with surface morphology complexity.Fig. 10 Mean cortical thickness vs age for the NCBR Yerkes dataset (N = 74). Comparison of gyrification index (GI) on the pial surface per hemisphere across the macaque (NMT, N = 29 (Seidlitz et al., 2018), chimpanzee (NCBR, N = 74), and human (MNI152-extended, N = 251 (Kötter et al., 2001)) datasets. Mean and standard deviation of GI increase with surface morphology complexity. Fig. 11Surface-based registration between the chimpanzee and human average brains. Illustration of the Chimpanzee Brainnetome Atlas (CBA) on the original Yerkes29 white cortical surface, mapped to the MNI-NCBR white cortical surface and to the human fsaverage white cortical surface. Excellent agreement of boundaries is observed across all three surfaces. (The brain surfaces across species have been resized to a common size for ease of visualization. The color map for the CBA has been reproduced from the original publication of Wang et al. (2025).)Fig. 11 Surface-based registration between the chimpanzee and human average brains. Illustration of the Chimpanzee Brainnetome Atlas (CBA) on the original Yerkes29 white cortical surface, mapped to the MNI-NCBR white cortical surface and to the human fsaverage white cortical surface. Excellent agreement of boundaries is observed across all three surfaces. (The brain surfaces across species have been resized to a common size for ease of visualization. The color map for the CBA has been reproduced from the original publication of Wang et al. (2025).) Discussion Validation remains a challenging endeavor for any newly developed processing pipeline, such as CIVET-Chimp, and, here specifically, the limited availability of MRI scans of chimpanzees has impeded on large-scale calibration of the pipeline. As such, comparisons with other in vivo and ex vivo datasets are not possible due to the lack of public data available for chimpanzees, other than the NCBR datasets, for which the relatively low contrast-to-noise ratio (CNR) of the MRI scans (see Fig. 1) raises concerns regarding the significance of the predicted cortical thickness measurements. In the current work, special attention has been given to manually correct the tissue classification, as required, to guard against topological errors in the cortical surfaces due to MR image artifacts (e.g. high intensity voxels representing blood vessels), but such annotations cannot overcome for lack of contrast in the final placement of the surfaces. Such annotations are provided via an input mask, without affecting the automation of the pipeline. The cortical thickness measurements predicted by CIVET-Chimp are compared to two previous independent studies by Demirci et al. (2023) and Xiang et al. (2020), both using the 3T scans of the NCBR dataset. In the first study, 54 subjects (19 males, 35 females; age 8.0-53.0yo) were processed using CAT12 and SPM12. The atrophy of the cortex with age predicted by CIVET-Chimp (Fig. 9) compares favorably well with that of Demirci et al. (2023) (Fig. 11), up to a constant shift of about 0.2 mm (CIVET-Chimp values being slightly smaller). In the second study, mean cortical thickness and pial surface area, obtained by a customized FreeSurfer pipeline, are reported for the ROIs defined by the Desikan-Killiany (D-K) atlas (see Xiang et al., 2020, Supplementary Materials Tables S1 and S2). For comparison with CIVET-Chimp, the human D-K atlas, defined on fsaverage, was transposed to the CIVET-Chimp average surface using the transformation obtained using surface-based registration (see Section 3, Fig. 11). It is noted that Xiang et al. (2020) used an older version of the D-K atlas with extra regions. Specifically, in the modern version of the D-K atlas, the regions for the temporal lobe pole and frontal lobe pole have been agglomerated to neighboring structures. Such a discrepancy will affect mostly surface area. The results, shown in Fig. 12, reveal a distinct trend of under-estimation of the cortical thickness across the brain for Xiang et al. (2020), in comparison to CIVET-Chimp and the Demirci et al. (2023) study, even when disregarding the two merged regions. However, the same global patterns across major lobes are clearly visible and follow the expected trends. Regional surface area shows a better agreement between CIVET-Chimp and the Xiang et al. (2020) study, except for the regions with redistributed area. This concordance also suggests that the D-K atlas on fsaverage has been accurately mapped from the human brain to the chimpanzee brain via the proposed surface-based registration approach. Fig. 12Mean cortical thickness and pial surface area by region of interest for the Desikan-Killiany (D-K) atlas for the 3T NCBR subjects. The D-K labels on the CIVET-Chimp average surface are obtained using the surface registration transformation to the fsaverage human brain. A comparison is made to the results of Xiang et al. (2020), for the same dataset (taken from Supplementary Materials Tables S1 and S2).Fig. 12 Mean cortical thickness and pial surface area by region of interest for the Desikan-Killiany (D-K) atlas for the 3T NCBR subjects. The D-K labels on the CIVET-Chimp average surface are obtained using the surface registration transformation to the fsaverage human brain. A comparison is made to the results of Xiang et al. (2020), for the same dataset (taken from Supplementary Materials Tables S1 and S2). Conclusions The CIVET-Chimp pipeline is a novel automated pipeline specifically tailored for processing structural T1-weighted chimpanzee MRI scans. It is a natural extension of the CIVET-macaque pipeline (Lepage et al., 2021) with species-specific templates, models, priors, and schedules. It can be used for registration, segmentation, masking, and cortical surface reconstruction providing surface-based morphometrics. The pipeline was also used to create an average surface template, derived from the cortical surfaces of the complete NCBR 3 T MRI dataset, representing, to date, the largest dataset used to generate a chimpanzee common surface template, unlike the previously available Yerkes29 surface composed of 29 female animals. Since the availability of previously collected scans of chimpanzees is limited, the ability to exploit these datasets using a tool such as CIVET-Chimp is even more valuable. The culmination of the CIVET pipeline suite allows the processing of human, macaque, and chimpanzee brains based on species-specific models and standardized spaces in a common environment, thus offering new opportunities for comparative and translational studies of neuroanatomy between primates. CIVET-Chimp is particularly well-suited for morphological analyses of cortical folding patterns. For instance, the morphological patterns and variability of the inferior frontal sulcus – a critical sulcal landmark below which lies the homologue of Broca’s region in the chimpanzee brain – were recently examined in the NCBR population using surface reconstructions generated with the current pipeline (Nolan et al., 2025). The average location and spatial variability of the sulcus were quantified in the form of spatial surface probability maps that were overlaid on the MNI-NCBR average surface template presented here. Given that a comparable morphological analysis of the inferior frontal sulcus was also performed in the human brain (Nolan et al., 2024), this allows cross-species comparisons and offers a unique opportunity to explore the evolutionary continuity and divergence of such neuroanatomical features. CRediT authorship contribution statement Claude Lepage: Writing – review & editing, Writing – original draft, Software, Methodology, Formal analysis, Conceptualization. Erika Nolan: Writing – review & editing, Writing – original draft, Data curation. Trisanna Sprung-Much: Writing – review & editing, Data curation. Gabriel A. Devenyi: Writing – review & editing, Resources, Methodology, Data curation. Michael Petrides: Writing – review & editing, Supervision. Alan C. Evans: Writing – review & editing, Supervision, Funding acquisition. Declaration of competing interest The authors report no competing interest.