Strengths and challenges of longitudinal non-human primate neuroimaging

Abstract

Longitudinal non-human primate neuroimaging has the potential to greatly enhance our understanding of primate brain structure and function. Here we describe its specific strengths, compared to both cross-sectional non-human primate neuroimaging and longitudinal human neuroimaging, but also its associated challenges. We elaborate on factors guiding the use of different analytical tools, subject-specific versus age-specific templates for analyses, and issues related to statistical power.

Keywords: Ageing; Development; Magnetic resonance imaging; Non-human primate; Simulation; Templates.

Fig. 1

Power simulation of longitudinal (L) and cross-sectional (X) designs in two different scenarios: (A) a developmental/ageing study, where age varies within or between subjects; and (B) an interventional study with one control group and one treatment group (with equal number of subjects in each group). In the interventional scenario, the cross-sectional design with two groups is compared to a longitudinal design with two groups scanned once before treatment and once or several times after treatment (to study the sustainability of the treatment effect). The increased power for longitudinal experiments is illustrated using asymptotic statistical methods for simulated data with a fixed effect size and a fixed detection threshold, using the longpower package (https://cran.r-project.org/web/packages/longpower/index.html). It should be noted that practically, signal variability is not homogenous across brain regions and that different power value estimates would likely be obtained for different brain regions (Suckling et al 2014).

Fig. 2

Quality control (QC) of T1w longitudinal images. A-D. Outlier identification. A. Brain surface estimation of an anesthetised macaque during five time points using PREEMACS Long. B. Estimation of intensity non-uniformity (INU) across acquisition points using MRIqc (Esteban et al., 2017) customized for the macaque by PREEMACS. C. Mean of the Brain Surface for the five acquisition points. The ROI defined by the dotted line corresponds to the precentral gyrus (pcg). Colour bar of cortical thickness (CT) in mm. D. CT estimation of pcg for each acquisition point. Point 3 is considered an outlier with a pcg CT and an INU that were statistically different from the other time points. E-H. Lack of longitudinal trends on QC metrics associated with head motion during awake scanning. E. Longitudinal scans of an awake macaque (one scan every 6 months) F. Median of INU field (INU med) as extracted by the N4ITK algorithm (values closer to 1.0 are better) across six acquisition points. G. Full-Width Half Maximum Smoothness (FWHM) of the spatial distribution of the image intensity values in units of voxels (lower values are better) for each time point. H. Entropy Focus Criterion (EFC) as a function of time points. EFC uses the Shannon entropy of voxel intensities as an indication of ghosting and blurring induced by head motion (lower values are better). The dotted line (F-H) corresponds to the best linear model between the QC metrics and the six acquisition points, with a slope close to zero for the three QC metrics, indicating no temporal trends in head motion.

Fig. 3

Anatomical measures such as Gray Matter Density (GMD), can change drastically across the lifespan, especially during early development (A) and late adulthood (C).

Fig. 4

Simulation results showing the difference between grey matter density estimates and the ground truth following registration to an age-specific template (blue) and a subject-specific template (orange) in a pipeline combining linear and non-linear registration steps.