MEBRAINS 1.0: A new population-based macaque atlas

Abstract

Due to their fundamental relevance, the number of anatomical macaque brain templates is constantly growing. Novel templates aim to alleviate limitations of previously published atlases and offer the foundation to integrate multiscale multimodal data. Typical limitations of existing templates include their reliance on one subject, their unimodality (usually only T1 or histological images), or lack of anatomical details. The MEBRAINS template overcomes these limitations by using a combination of T1 and T2 images, from the same 10 animals (Macaca mulatta), which are averaged by the multi-brain toolbox for diffeomorphic registration and segmentation. The resulting volumetric T1 and T2 templates are supplemented with high-quality white and gray matter surfaces built with FreeSurfer. Human-curated segmentations of pial surface, the white/gray matter interface, and major subcortical nuclei were used to analyze the relative quality of the MEBRAINS template. Additionally, 9 computed tomography (CT) scans of the same monkeys were registered to the T1 modality and co-registered to the template. Through its main features (multi-subject, multimodal, volume-and-surface, traditional, and deep learning-based segmentations), MEBRAINS aims to improve integration of multimodal multi-scale macaque data and is quantitatively equal to, or better than, currently widely used macaque templates. We provide a detailed description of the algorithms/methods used to create the template aiming to furnish future researchers with a map-like perspective which should facilitate identification of an optimal pipeline for the task they have at hand. Finally, recently published 3D maps of the macaque inferior parietal lobe, (pre)motor and prefrontal cortex were warped to the MEBRAINS surface template, thus populating it with a parcellation scheme based on cyto- and receptor architectonic analyses. The template is integrated in the EBRAINS and Scalable Brain Atlas web-based infrastructures, each of which comes with its own suite of spatial registration tools.

Keywords: T1- and T2-weighted MRI; brain atlas; computed tomography skull imaging; group template; macaque brain.

Fig. 1.

Number of publications per year related to brain templates in macaque monkeys. A PubMed search query was performed November 2023 using the following keyword combination: (“monkey” OR “macaque” OR “NHP” OR “non-human primate”) AND (“template” OR “atlas”) AND (brain). Polynomial fit with R² = 0.8411.

Fig. 2.

Overview of the pipeline used for the generation of a population-based template that represents an average of high-resolution structural T1 and T2 MRI scans as well as CT. Panels A–D highlight the four major processing blocks of which the pipeline is composed.

Fig. 3.

Overview of the MB model (adapted from Fig. 1 of Brudfors et al., 2020). (A) Distribution of brains (T1 and T2 images; monkeys X₁,…X_N). (B(i)) Group-wise learning of the optimal population average. (B(ii)) Pairwise deformations obtained by composing deformations via the optimal population average. Here, application of ψ_i^-1 and of ψ_j enables registration of brain x_i to the optimal population average, and of the optimal population average to brain x_j, respectively, whereas use of ψ_i^-1 + ψ_j results in the registration of brain x_i to brain x_j.

Fig. 4.

Three orthogonal sections of the volumetric MEBRAINS_T1 (A) and MEBRAINS_T2 (B) templates. The NIFTI-volumes used to create this figure can be found in Supplementary Material, and are also made publicly available via the EBRAINS platform (https://doi.org/10.25493/VS6E-7KR).

Fig. 5.

Three orthogonal sections (A–C) and 3D rendering (D) of the volumetric MEBRAINS_CT template. The corresponding NIFTI-volume can be found in the Supplementary Material, and is also made publicly available via the EBRAINS platform (https://doi.org/10.25493/VS6E-7KR).

Fig. 6.

Three orthogonal sections of the ANTS10 templates generated from T1 (A) and T2 (B) images. To facilitate comparison with the corresponding MEBRAINS templates, the levels shown are the same as those depicted in Figure 4.

Fig. 7.

Generation of pial and white matter surfaces using MB (A and B) and postprocessing with FreeSurfer (C). (A) White matter mask overlaid on the MEBRAINS_T1 template. (B) Gray matter mask overlaid on the MEBRAINS_T1 template. (C) Pial (magenta) and white matter (yellow) boundaries overlaid on the MEBRAINS_T1 template. The sagittal, coronal, and horizontal sections depicted correspond to coordinates x13, y0, and z4, respectively.

Fig. 8.

MEBRAINS surface templates representing the pial (A) and white matter (B) brain surfaces as well that of the skull (C). The corresponding gifti files can be found in the Supplementary Material, and are also made publicly available via the EBRAINS platform (https://doi.org/10.25493/VS6E-7KR).

Fig. 9.

(A) Human-curated segmentation of the cortical ribbon, white matter and lateral ventricles, as well as of diverse subcortical nuclei, and the anterior commissure. (B, C, and D) Areas of the macaque inferior parietal lobule (Niu et al., 2021) and of the motor and pre-motor cortex (Rapan et al., 2021) warped from the Yerkes19 template to MEBRAINS. Areas are overlaid on the folded surface of MEBRAINS in (B), the flat maps in (C), and exemplary sections of MEBRAINS_T1 are shown in (D). Abbreviations: 4a, 4p = primary motor areas 4a and 4p; AC = anterior commissure; Am = amygdala; Cx = cerebral cortex; Cd = caudate nucleus; Cl = claustrum; F2d, F2v = dorsal and ventral parts of dorsolateral premotor area F2; F3 = supplementary motor area F3; F4d, F4s, F4v = dorsal, sulcal and ventral parts of lateral premotor area F4; F5d, F5v = dorsal and ventral parts of ventrolateral premotor area F5; GP = globus pallidus; LV = lateral ventricle; Opt = parietal area Opt; PF = parietal area PF; PFop = parietal area PFop; Pu = putamen. The sagittal, coronal, and horizontal sections depicted in (A) and (D) correspond to coordinates x13, y0, and z4, respectively.

Fig. 10.

Eight of commonly used rhesus macaque brain templates (NMT v2.0 (Seidlitz et al., 2018), Yerkes19 (Donahue et al., 2018; Van Essen et al., 2012), D99 (Reveley et al., 2017), MNI (Frey et al., 2011), F99 (Van Essen, 2004), INIA19 (Rohlfing et al., 2012), ONPRC18 (Weiss et al., 2021), and 112RM-SL (McLaren et al., 2009)), as well as our ANTS10_T1 volume (i.e., the template built with ANTS using the same 10 datasets as MEBRAINS_T1) were registered to MEBRAINS using ANTS. The meta-template represents the average of all these datasets with the exception of 112RM-SL.

Fig. 11.

Pearson correlation and “1 – Focal Entropy Difference” (scaled to facilitate comparisons with the correlation method: 0 – total dissimilarity; 1 – total similarity) calculated for the reference anatomy MEBRAINS compared with the following templates: MEMRAINS, ANTS10_T1, NMT v2.0, Yerkes19, D99, MNI, F99, INIA19, ONPRC18, and 112RM-SL. Comparison of MEBRAINS with itself (value 1) provides the reference for the ideal registration.

Fig. 12.

D99 atlas registered to MEBRAINS using the MB toolbox (A), ANTS (B), and “run-N-select-high-probability-values” (C) approaches. The different registrations of the atlas are overlaid on the MEBRAINS template.

Fig. 13.

Masking performance of the U-net convolutional neural network using one example model. The predicted mask at the end of the training for an “easy” anatomy (A) and a “difficult” anatomy (C), and the dice score during the training (B). The performance for the “difficult” anatomy (red line in B) reached the optimal performance later than for the “easy” anatomy (green line in B). The maximum average dice score is 0.9887, and was reached in epoch 38.

Fig. 14.

(A) Anatomies of the six templates used to quantitatively compare the quality of the EBRAINS template. (B) Structures that were selected for the MEBRAINS_T1 template: Am = amygdala; Cd = caudate; Cl = claustrum; GM = cortical gray matter; Pu = putamen; WM = white matter.

Fig. 15.

C2N parameter distribution of means for the templates shown in Table 2 and Figure 14A. Parameters were calculated for the six selected sub-structures separately, and numbers represent the median values.

Fig. 16.

KI parameter distribution of means for the templates shown in Table 3 and Figure 14. Parameters were calculated for the six selected sub-structures separately, and numbers represent the median values.