(E-182) Non-Negative Matrix Factorization of Brain Tissue Stiffness Images from Magnetic Resonance Elastography

Introduction:: Computational models are an important tool in traumatic brain injury (TBI) research, as the brain’s response to a skull acceleration cannot typically be studied in vivo [1]. Magnetic resonance elastography (MRE) is a noninvasive imaging technique capable of measuring precise mechanical properties of the brain. Material properties obtained through this technique can provide TBI models with accurate input parameters. Current computational models are based on average property maps acquired from a wide range of subjects that do not capture the individual variability between subjects. More recent studies are working to develop subject-specific models [2,3], but these incur the opposite problem of being too specific and do not accurately capture the spread of material properties across the population. Here we seek to describe the population variability in brain mechanical properties in a tractable, low-dimensional form [4]. Non-negative matrix factorization (NMF) can be used to decompose this data into factors and calculate loadings to determine contributions to the average from different regions of the brain. NMF has been shown to capture patterns of neuroanatomical variation and this study will expand that approach to study differences in mechanical properties [5,6]. Individual factors correspond to regions that co-vary between individuals, and we examine which of these factors is most affected by age, known to reduce brain tissue stiffness.


Materials and Methods:: MRE data was collected using a Siemens 3T Prisma scanner with a 64-channel head/neck coil. Vibrations were generated using a Resoundant pneumatic actuation system delivered through a soft pillow driver at 50 Hz. A 3D multiband, multishot spiral sequence allows for data collection with an OSCILLATE [7] encoding scheme to accelerate acquisition time. Data was acquired at 1.5 mm isotropic resolution in less than 5 minutes. A cohort of 65 healthy subjects were recruited for this study, ranging in age from 14-80 years old. Images were reconstructed using PowerGrid [8] and post-processed to obtain maps of complex motion with sensitization in the x, y, and z directions. A nonlinear inversion (NLI) algorithm [9] was used to estimate material properties of the brain tissue including shear stiffness. All subject data was registered to standard space using the MNI-152 atlas. Maps of shear stiffness for all subjects were input into a matrix which was then decomposed to determine contributions from different unconfined regions of the brain. An analysis of reconstruction error was conducted to determine the optimal number of factors for further evaluation. Calculated loadings can then be used to determine correlations between age and the material properties at each region.


Results, Conclusions, and Discussions:: NMF produced factors that appeared to capture different anatomical regions, that co-vary in mechanical stiffness across the population. Figure 1A shows the resulting regions from each factor in 4 different slices of the brain. Factor 1 appears to be mostly the frontal and temporal lobes of the brain whereas Factor 2 includes the occipital and parietal lobes. The rest of the factors also tend to correspond with common brain regions such as white matter tracts (3, 5, 7), ventricles (4), subcortical structures (6), and the cerebellum (8). Figure 1B shows how loadings across subjects for given factors were associated with age. While all regions decreased with age, as expected, some were more strongly correlated. Stiffness values associated with the frontal/temporal lobes as well as periventricular white matter are highly correlated with age whereas the cerebellum and ventricles remain more constant over time, consistent with previous MRE aging research that found the ventricles and cerebellum are not expected to vary at all. This technique provides a novel approach for studying regional variation in the structural significance of brain tissue, and its relation to aging, without being confined to the predetermined regions-of-interest of the brain that are often used in this type of analysis.

Matrix factorization will allow researchers to decompose the image space to determine contributions from specific brain regions and provide more precise data for models of injury mechanics. Future work will include the incorporation of additional data into the matrix decomposition including other material properties, like damping ratio, or MRE data acquired from different actuation directions and frequencies. This technique will provide TBI models with more precise input parameters to better study the importance of regional differences in the analysis of brain biomechanics.


Acknowledgements (Optional): : The project was supported by NIH grant U01-NS112120.

References (Optional): :

1. Bayly, P. V. et al. MR Imaging of Human Brain Mechanics In Vivo: New Measurements to Facilitate the Development of Computational Models of Brain Injury. Ann. Biomed. Eng. 2021 4910 49, 2677–2692 (2021).


2. Upadhyay, K. et al. Development and validation of subject-specific 3D human head models based on a nonlinear visco-hyperelastic constitutive framework. J. R. Soc. Interface 19, (2022).


3. Alshareef, A. et al. Integrating material properties from magnetic resonance elastography into subject-specific computational models for the human brain. Brain Multiphysics 2, 100038 (2021).


4. Lam, F., Li, Y. & Peng, X. Constrained Magnetic Resonance Spectroscopic Imaging by Learning Nonlinear Low-Dimensional Models. IEEE Trans. Med. Imaging 39, 545–555 (2020).


5. Yang, Z. & Oja, E. Linear and nonlinear projective nonnegative matrix factorization. IEEE Trans. Neural Networks 21, 734–749 (2010).


6. Sotiras, A., Resnick, S. M. & Davatzikos, C. Finding Imaging Patterns of Structural Covariance via Non-Negative Matrix Factorization. Neuroimage 108, 1 (2015).


7. McIlvain, G. et al. OSCILLATE: A low-rank approach for accelerated magnetic resonance elastography. (2022) doi:10.1002/mrm.29308.


8. Cerjanic, A. et al. PowerGrid: A open source library for accelerated iterative magnetic resonance image reconstruction. Proc. Intl. Soc. Mag. Res. Med. 24 525 (2016).


9. McGarry, M. D. J. et al. Multiresolution MR elastography using nonlinear inversion. Med. Phys. 39, 6388–6396 (2012).