Characterizing Heterogeneity in Brain Morphology in Traumatic Brain Injury Using Normative Modeling

Traumatic brain injury (TBI) is heterogeneous, complicating efforts to develop standardized diagnostic and therapeutic approaches. Conventional group-based analyses often obscure individual differences by averaging across diverse injury patterns. By contrast, by comparing deviation from the expected distribution observed in healthy controls, normative modeling can capture patient-specific deviations from expected neuroanatomical norms. The aim of this research was to use normative modeling to capture individual variability in brain morphometry on individuals who have experienced a TBI.

In this study, we applied traditional case-control and a normative modeling approach to cortical and subcortical MRI. Data were analyzed from the Enhancing NeuroImaging Genetics through Meta-Analysis Consortium Adult Moderate-to-Severe TBI working group. Primary outcome measures were cortical thickness and subcortical volumes, derived from the Destrieux and Freesurfer subcortical atlases. Case-control tests were conducted using linear models controlling for age, sex, site, and intracranial volume. Normative modeling was carried out by accessing the Predictive Clinical Neuroscience Portal, estimating individual deviations using a model of Bayesian linear regression with likelihood warping.

A total of 631 (407 patients with TBI, 224 healthy controls) MR images were processed. Conventional case-control analyses identified significant group differences in 153 regions of possible 178 cortical or subcortical regions. Normative modeling revealed far greater heterogeneity. No more than 23% of patients with TBI shared an extreme deviation (z-score >2 or z < -2) in the same region, but every region experienced at least 1 extreme positive or negative deviation across individuals. Stratifying by Glasgow Coma Scale (GCS) severity sustained this pattern; even within GCS 13-15, GCS 9-12, and GCS 3-8 subgroups, regional convergence did not exceed 34%. The median number of deviations increased with injury severity (GCS 13-15 = 9, GCS 9-12 = 19, GCS 3-8 = 22), demonstrating that group averages mask highly individualized morphologic abnormalities as injury severity increases.

Normative modeling detects participant-specific cortical and subcortical abnormalities that conventional group comparisons overlook, better representing the true diversity of TBI-related morphologic changes. Generating individualized "morphologic fingerprints" may ultimately advance prognostic accuracy and lay the foundation for personalized interventions in research and clinical practice.