White matter brain charts: modelling diffusion imaging data across the lifespan

1276 

Abstract Submission 

Ramona Cirstian¹, Natalie Forde², Christian Beckmann³, Andre Marquand⁴

¹Donders Institute for Brain, Cognition and Behaviour, Nijmegen, Gelderland, ²Donders Institute for Brain, Cognition, and Behavior, Radboud University Nijmegen, Nijmegen, NL, Nijmegeb, Netherlands, ³Radboud University Medical Centre, Donders Institute for Brain, Cognition and Behaviour, Nijmegen, Gelderland, ⁴Radboud University Nijmegen, Nijmegen, Gelderland

Ramona Cirstian  
Donders Institute for Brain, Cognition and Behaviour
Nijmegen, Gelderland

Natalie Forde  
Donders Institute for Brain, Cognition, and Behavior, Radboud University Nijmegen, Nijmegen, NL
Nijmegeb, Netherlands

Christian Beckmann  
Radboud University Medical Centre, Donders Institute for Brain, Cognition and Behaviour
Nijmegen, Gelderland

Andre Marquand  
Radboud University Nijmegen
Nijmegen, Gelderland

Diffusion MRI (dMRI) is a neuroimaging modality used to evaluate brain structure at a microscopic level and can be exploited to map the direction of the white matter fibers in the brain. In this study, we curated a reference cohort comprising high-quality multi-shell diffusion neuroimaging data from 10 sites across 5 different studies (N=25104; ages 2–100) and using normative modeling, we delineated lifespan trajectories of fractional anisotropy and mean diffusivity for 50 white matter tracts based on the JHU white matter tract atlas. Normative modelling is an innovative method used to model biological and behavioral variation across a study population and can be used to make statistical inferences at an individual level [1]. This is achieved by mapping a response variable (e.g., neuroimaging derived phenotypes) to a covariate (e.g., age) in a similar way growth charts are used in pediatric medicine to map the height or weight of children to their age. Establishing reference models to capture population variations and examining individual deviations is important for comprehending inter-individual variability and its connection to the onset and advancement of medical conditions [2].

We compiled data from four large open-source datasets for this analysis: UK Biobank (N = 22654, age 46-82), HCP Young Adult (N = 1065, age 22-37), HCP Development (N = 454, age 8-21), HCP Aging (N = 239, 35-100) and Developing HCP (age 2-3). The data were pre-processed using the HCP pipeline with the exception of the UK Biobank which was obtained already pre-processed [3]. The data were later processed using DTIfit and TBSS to obtain the FA and MD means for each tract using the JHU atlas.
A normative model was trained with the training set (n=16736) to estimate the normal range of each IDPs value according to age. To account for the possible non-linear effects and non-Gaussian distributions within the dataset, we used a warped Bayesian linear regression (BLR) model [4]. We used fixed effects to model the effect of site as demonstrated in our previous publications [5]. Next, the test set was used to estimate each subjects' deviation from the normal range of each IDP by computing the individual z-score (equation 2). The fit statistics of the model were computed including explained variance, skew, and kurtosis. The deviation for each subject was visualized by plotting the individual z-scores across the mean and centiles of variation predicted by the model.

The initial concatenation of the five datasets showed severe site effects caused by discrepancies between scanner parameters, resulting in data scaling. These effects are visualized in Figure 1 and 2 on the left side of each panel. Figure 1A displays the scatterplot of average FA values within the body of the corpus callosum obtained from the skeletonized data. Each dataset is color-coded to emphasize the discrepancy in data distribution and scaling. The same principle applies to Figure 1C, which represents the average FA values within the uncinate fasciculus right (on the skeleton). In contrast, on the right-hand side in Figure 1C and 1D, the datapoints represent the z-scores obtained from applying the BLR normative modeling, which predicts the centiles of variation across the entire lifespan. This approach accounts for the non-Gaussianity of the dataset while adjusting and rescaling the datapoints to correct for site effects. The same principles are applied to Figure 2, where the MD values for the same two tracts are represented.

In this study, we developed normative brain charts for fractional anisotropy (FA) and mean diffusivity (MD) using a diverse lifespan dataset. Our models, adaptable to non-Gaussian distributions, accommodate new sites and prioritize individual variations, moving beyond group-level inferences for personalized insights.

Aging

Lifespan Development Other ¹

Bayesian Modeling ²

Classification and Predictive Modeling

Diffusion MRI Modeling and Analysis

Aging

Computational Neuroscience

Data analysis

Development

Modeling

Statistical Methods

Tractography

WHITE MATTER IMAGING - DTI, HARDI, DSI, ETC