Charting brain growth and aging at high spatial precision

Abstract

Defining reference models for population variation, and the ability to study individual deviations is essential for understanding inter-individual variability and its relation to the onset and progression of medical conditions. In this work, we assembled a reference cohort of neuroimaging data from 82 sites (N=58,836; ages 2-100) and used normative modeling to characterize lifespan trajectories of cortical thickness and subcortical volume. Models are validated against a manually quality checked subset (N=24,354) and we provide an interface for transferring to new data sources. We showcase the clinical value by applying the models to a transdiagnostic psychiatric sample (N=1985), showing they can be used to quantify variability underlying multiple disorders whilst also refining case-control inferences. These models will be augmented with additional samples and imaging modalities as they become available. This provides a common reference platform to bind results from different studies and ultimately paves the way for personalized clinical decision-making.

Keywords: big data; brain chart; growth chart; human; individual prediction; lifespan; neuroscience; normative model.

Figure 1.. Normative model overview.

(A) Age density distribution (x-axis) of each site (y-axis) in the full model train and test, clinical, and transfer validation set. (B) Age count distribution of the full sample (N=58,836). (C, D) Examples of lifespan trajectories of brain regions. Age is shown on x-axis and predicted thickness (or volume) values are on the y-axis. Centiles of variation are plotted for each region. In (C), we show that sex differences between females (red) and males (blue) are most pronounced when modeling large-scale features such as mean cortical thickness across the entire cortex or total gray matter volume. These sex differences manifest as a shift in the mean in that the shape of these trajectories is the same for both sexes, as determined by sensitivity analyses where separate normative models were estimated for each sex. The explained variance (in the full test set) of the whole cortex and subcortex is highlighted inside the circle of (D). All plots within the circle share the same color scale. Visualizations for all ROI trajectories modeled are shared on GitHub for users that wish to explore regions not shown in this figure.

Figure 2.. Normative modeling in clinical cohorts.

Reference brain charts were transferred to several clinical samples (described in (A)). Patterns of extreme deviations were summarized for each clinical group and compared to matched control groups (from the same sites). (B) Shows extreme positive deviations (thicker/larger than expected) and (C) shows the extreme negative deviation (thinner/smaller than expected) patterns. (D) Shows the significant (FDR corrected p<0.05) results of classical case-control methods (mass-univariate t-tests) on the true cortical thickness data (top row) and on the deviations scores (bottom row). There is unique information added by each approach which becomes evident when noticing the maps in (B–D) are not identical. ADHD, attention-deficit hyperactive disorder; ASD, autism spectrum disorder; BD, bipolar disorder; EP, early psychosis; FDR, false discovery rate; MDD, major depressive disorder; SZ, schizophrenia.

Figure 3.. Evaluation metrics across all test sets.

The distribution of evaluation metrics in four different test sets (full, mQC, patients, and transfer, see Materials and methods) separated into left and right hemispheres and subcortical regions, with the skew and excess kurtosis being measures that depict the accuracy of the estimated shape of the model, ideally both would be around zero. Note that kurtosis is highly sensitive to outlying samples. Overall, these models show that the models fit well in term of central tendency and variance (explained variance and MSLL) and model the shape of the distribution well in most regions (skew and kurtosis). Code and sample data for transferring these models to new sites not included in training is shared.

Figure 3—figure supplement 1.. Comparison of the explained variance in cortical thickness across the different test sets.

The patterns appear to be robust and consistent across the different test sets.

Figure 3—figure supplement 2.. Showing the explained variance for each brain region across 10 randomized resampling of the full control test set.

Figure 3—figure supplement 3.. Per site explained variance across the different test sets.

Author response image 1.. Example of the online interactive visualizations created to help interpret the evaluation metrics.

This interactive figure was created for each evaluation metric (EV, MSLL, skew, and kurtosis) and all test sets (full controls, mQC controls, clinical, and transfer).