Analytic Bounds on GAMLSS Model Variability of Normative White Matter Brain Charts

Area of Science:

• Neuroimaging
• Developmental neuroscience
• Biostatistics

Background:

• Brain charts provide normative models for quantitative neuroimaging measures, aiding in identifying developmental trajectories and abnormalities.
• Recent advancements have extended brain charts to include microstructural and macrostructural white matter features.
• Assessing the variance of these brain charts is crucial for ensuring the stability of the underlying models.

Purpose of the Study:

• To implement and validate an analytic approach for characterizing parameter variability in generalized additive models for location, scale, and shape (GAMLSS) brain charts.
• To empirically validate the accuracy of the analytic variability assessment against a bootstrapping approach across a wide age range.
• To compare the stability estimates derived from analytic and empirical methods for white matter brain chart features.

Main Methods:

• An analytic approach was developed to characterize the variability of parameters in existing GAMLSS brain charts.
• Empirical validation was performed using a bootstrapping approach, comparing results to the analytic method from 0.2 to 90 years of age.
• The coefficient of variation (COV) was calculated for both analytic and empirical assessments across various white matter features.

Main Results:

• Analytic coefficient of variation (COV) remained below 5% for ages above 0.25 years; the maximum empirical COV was 7% at 0.2 years.
• High agreement was observed between analytic and empirical assessments, with a Pearson correlation of 0.776 for COV estimates across the lifespan.
• Volume and surface area showed the largest average COV for most white matter tracts, while analytic and empirical methods differed on features with the smallest COV (axial diffusivity vs. average length).

Conclusions:

• The analytic approach may overestimate model stability in white matter brain charts when the coefficient of variation is low.
• The empirical validation method is suitable for assessing the stability of GAMLSS models used in brain charting.
• Understanding variability is key to refining normative models for brain development and detecting abnormalities.