Poster R9 - Multifrequency Atlas of Human Brain Stiffness with Reduced Dimensionality

Introduction: Traumatic brain injury (TBI) is a leading cause of death, affecting 10 million people each year (Hyder, 2007). Because the mechanism of injury is impossible to study in vivo, previous research on TBI has been conducted using animal or cadaver models. However, these methods are limited in their similarity to the human brain. More recently, computational models of brain biomechanics have provided a method of quantifying brain displacement in response to simulated head impacts for understanding injury mechanisms. These models, however, require precise input parameters for model accuracy, including the mechanical properties of the brain. Previous work by Hiscox et. al. (2021) established a human brain atlas based on the mechanical properties of brain tissue measured with magnetic resonance elastography (MRE). Incorporating data from this atlas into computational models improved model performance and validation (Giudice, 2021). Here, we aim to build on the previous atlas by incorporating multifrequency MRE data across a wide age range of subjects. Brain tissue is known to be frequency-dependent, therefore the addition of multifrequency data will allow for better informed models. As the resulting data will contain voxel-wise mechanical property data for many subjects each at 3 actuation frequencies, this analysis becomes extremely high-dimensional, making it difficult to translate into a useful format for modeling. A further objective of this work was to reduce the dimensionality of this dataset and determine which unconfined regions of the brain covary in mechanical properties across the population.

Materials and

Methods: This study included 86 healthy subjects ranging in age from 14-78 years old (38M/48F). All data was collected using a Siemens 3T Prisma scanner with a 64-channel head/neck coil. MRE is a noninvasive technique used to measure induced tissue displacement through phase contrast imaging. Vibrations were applied to the head using a Resoundant pneumatic actuation system delivered through a soft pillow driver at 30, 50, and 70 Hz. A 3D multiband, multishot spiral sequence with OSCILLATE acceleration (Johnson, 2016; McIlvain, 2022) allowed for data collection in under 5 minutes per frequency. A nonlinear inversion algorithm (McGarry, 2012) was used to estimate material properties of the brain tissue from the displacement fields, specifically shear stiffness. All data is available at nitrc.org/projects/bbir (Bayly, 2021).
A 0.9 mm isotropic resolution T1-weighted anatomical scan was collected for each subject to aid in image registration. MRE magnitude images were registered to the standard space using the MNI-152 atlas. Resulting warp transformations were applied to the maps of shear stiffness at each actuation frequency. Voxel-wise stiffness data for all subjects at each actuation frequency were combined into a single matrix, X, which was decomposed using an orthogonal projective non-negative matrix factorization (NMF) to estimate both spatial factors, W, and loadings, H, shown in Figure 1 (Sotiras, 2015). We performed NMF using 8 factors and reconstructed the data matrix as X = WH. Resulting spatial components (W) were transformed into regional binary masks where we calculated the average stiffness values across the population in each region.

Results, Conclusions, and Discussions: The NMF produced 8 spatial factors that appeared to capture identifiable anatomical regions in the brain which co-vary in shear stiffness across the population (Figure 2). These spatial components can be combined to exhibit full-brain coverage with minimal overlap. The cortex of the brain was captured by Factor 2 (frontal/parietal), Factor 3 (midline), and Factor 4 (temporal). The remaining factors correspond to common brain regions such as the ventricles (1), white matter tracts (5 & 6), subcortical structures (7), and the cerebellum (8). This data-driven technique provides an innovative approach for studying regional variation in the mechanical behavior of brain tissue, without being confined to predetermined regions of interest based on anatomy. Average stiffness values across the population were determined for each spatial component. As expected, there was an overall increase in shear stiffness at higher frequencies, but this tends to vary by factor, as shown in Figure 3. At higher frequencies, the fluid-filled ventricles appear stiffer than surrounding brain tissue, which may result from 70 Hz having less wave penetration in this region. Future work will involve removing the ventricles from the images prior to running NMF to reduce this error.

Computational models of TBI not only rely on precise input parameters, but also large quantities of data to represent variability across the population. MRE provides important mechanical property information though it is a relatively new technique with limited data availability. Matrix factorization enables the decomposition of image space to determine contributions from specific brain regions, allowing for standardization across protocols and overall decreased variability. Reducing data in this way and applying it to various MRE studies will allow us to define standard mechanical properties of the brain for a given subject population. The incorporation of this data in an improved atlas of viscoelastic properties will provide a clear baseline of brain mechanics.

Acknowledgements (Optional): This project was supported by NIH grant U01-NS112120, and Office of Naval Research grant N00014-22-1-2198.