Date of Award

Summer 2025

Document Type

Open Access Thesis

Department

Biomedical Engineering

First Advisor

Ahmed Alshareef

Abstract

Traumatic brain injury (TBI) is a global health concern, with over 69 million cases annually. TBI risk depends on subject-specific neuroanatomical variation, yet research often only considers generic finite element (FE) models, based on the geometry and material composition of a 50th percentile male head. Although highly accurate, subject-specific models require expensive magnetic resonance (MR) imaging, elastography, and time-consuming computational resources. There is a need to create subject-specific TBI models that account for subject-specific variations, without neuroimaging. This study aims to create mathematical models to predict brain morphometry (dimensions, volume) [Objective 1] and mechanical properties (shear stiffness, damping ratio, octahedral shear strain) [Objective 2] from anthropometric measurements (age, sex, head dimensions). Such predictions could be used to scale generic TBI models, improving model accuracy.

Neuroimaging data, including MR structural imaging and elastography, were obtained from an open-source dataset. The full cohort contained 156 subjects (74 males, 82 females), aged 14-75, with neuroimages and associated metadata (age, sex, weight, height). Regional and total brain volume, head measurements, and brain dimensions were extracted from T1- and T2-weighted neuroimages and deep-learning brain segmentations (Brain Mask, SLANT-CRUISE) using ITK-SNAP (Version 3.8.0) and custom MATLAB code (Version R2024b). Brain mechanical properties (50th, 95th percentile shear stiffness (SS), damping ratio (DR), octahedral shear stiffness (OSS)) were extracted from whole-brain MR elastography data (30, 50, 70Hz) from a subset of 85 subjects (37 males, 48 females). Simple linear (SLR) and stepwise multiple linear (MLR) regression models were created to predict brain measurements and mechanical properties from head anthropometry. Principal component analysis (PCA) was used to identify brain shape modes that explain 95% of the variation. Principal component regression (PCR) was used to relate brain shape to mechanical properties.

Analysis of the SLR models indicated that head width, length, depth, and perimeter are moderate predictors of respective brain width, length, depth, and volume (R2: 0.61, 0.73, 0.69, 0.64, respectively). The MLR models improved R2 values for brain width, length, and volume by 19%, 4%, and 22%, respectively, suggesting multivariable dependence of brain geometry on combinations of anthropometric predictors. The SLR models for mechanical properties performed poorly, with low R2 values when predicting SS, DR, and OSS from age only (R2,50th: 0.33, 0.31, 0.00; R2,95th: 0.11, 0.32, 0.06), and the MLR models improved R2 values by including combinations of head dimensions and sex in the regression equations. The high error in mechanical property predictions suggests a dependence on factors beyond head anthropometry. The results of PCA showed that five principal components explained 95% of brain shape variation, capturing frontal-occipital regional shape variation (40.8%), diagonal hemispheric asymmetry (21.3%), sagittal hemispheric asymmetry (15.3%), and a width-length relationship (5.3%). PCR models using these components to predict mechanical properties showed large deviations from true values, suggesting that brain shape alone is insufficient for characterizing tissue mechanics.

Rights

© 2025, Jessica Restivo

Download

Share

COinS