Monday  |  Conference Center A  |  10:00 AM–10:20 AM

#13812, A Machine Learning Approach to Study the Effect of Human Head Shape on the Risk of Traumatic Brain Injury

Traumatic brain injuries are a host of neurological disorders caused by the rapid application of external forces on the head. Computational head models are commonly used to simulate the mechanical response of brain during such events, which are then correlated to the risk of injury via strain-based metrics. Despite the significant variability in head shape across the human population, existing head models are composed of an “average” head geometry. A natural question thus arises: what is the effect of head geometry on the strain-response of brain? A quantitative investigation into this question is a key step toward improving the confidence on the injury risk predictions of existing models, and can provide critical insight into the relative risk of injury among different people. In this study, a machine learning approach is developed to study the effect of head shape on the strain response of brain under rapid rotational accelerations, which is correlated to injury risk.

Anatomical brain images of 25 healthy volunteers (age: 21-49 y, 12 male and 13 female) are acquired using magnetic resonance imaging. These images are affinely registered to the standard MNI152 (Montreal Neurological Institute) brain image, resulting in transformation matrices that quantify head shape. A data-driven method based on multi-dimensional Gaussian kernel-density estimation and diffusion maps is then used to generate 100 realizations of the transformation matrix directly from the available data set of 25 subjects. 3D head models are developed for the 100 head geometries, for which visco-hyperelastic material properties of different brain regions are calibrated from experimental stress-strain data in the literature. A rapid rotational acceleration is applied to the head model about the inferior/superior axis, and resulting strain fields are used to compute three common injury predictors: brain volume fraction with strain above an injury threshold, peak and average strains. Two strain measures are considered: maximum axonal strain, and cumulative maximum principal strain. A Gaussian-process surrogate model is trained to create a low-cost mapping between transformation matrices and injury predictors, which is then used in the framework of Monte-Carlo simulations to obtain injury predictors for 10000 new input realizations. Results suggest that head shape has a considerable influence on the spatial distribution of strain, and larger heads are associated with a greater injury risk.

Kshitiz Upadhyay   Johns Hopkins University
Roshan Jagani   Johns Hopkins University
Dimitris Giovanis   Johns Hopkins University
Ahmed Alshareef   Johns Hopkins University
K.T. Ramesh   Johns Hopkins University