Title: Optimizing the computational modeling of traumatic brain injury with machine learning and large animal modeling

Abstract: Legislation across all 50 states in the U.S. addresses sports-related mild traumatic brain injury (mTBI), requiring medical clearance before students can return to play. However, there currently lacks an objective, unbiased method to pre-screen potential mTBI sufferers and diagnose mTBI. While imaging holds promise as an objective diagnostic tool, it is expensive and logistically challenging. Wearable devices that monitor head impacts offer a promising pre-screening method for individuals susceptible to mTBI, while biomechanics computation can link these wearables to imaging and mTBI pathologies. This dissertation advances TBI biomechanics modeling by integrating machine learning techniques with large animal models, enhancing the precision and applicability of biomechanical modeling for improved TBI risk assessment.

The computational biomechanics modeling of TBI typically involves the sequence of head impact, head movement kinematics, brain deformation, and resulting injuries. Traditional methods encounter challenges such as imprecise kinematic measurements in humans, time-intensive modeling processes, and limited generalizability across various types of head impacts. To address these limitations, my research leverages extensive simulated and real-world head impact data collected at CamLab to optimize the modeling process. To reduce the time consumption, machine learning head models have been developed to rapidly compute the whole-brain strain from head kinematics. To improve the accuracy, deep learning-based models have been employed to denoise kinematic measurements obtained from wearable sensors. Additionally, transfer learning and unsupervised domain adaptation techniques have been utilized to generalize the machine learning head models to diverse types of head impacts. Furthermore, to bridge the gap between biomechanics and medical imaging for enhanced mild TBI diagnosis, a novel impact porcine model has been devised to establish connections between biomechanics, neuroimaging, and histopathology.

Join me for my defense presentation, where I will delve into how this refined modeling approach not only advances our comprehension of TBI mechanics but also lays the groundwork for dependable, non-invasive diagnostics and early interventions in sports and beyond.