Monitoring brain strain

Brain exposure data from real athletes, across whole matches, seasons, and careers, is what medics, coaches, and sports governing bodies need but rarely get cleanly. Instrumented mouthguards can capture head motion at the moment of impact, but translating that motion into what’s actually happening inside the brain is much harder: a full finite element simulation gives a detailed picture of brain strain, but can take 5–6 hours per impact, far too slow to use pitch-side, or across a full season of recordings.

We collaborate with sport governing bodies and mouthguard manufacturers to access thousands of head-impact recordings, capturing translational and rotational head motion in contact sport. During her PhD in the HEAD Lab, Dr Emily Chan developed machine-learning surrogate models that learn the relationship between this head-impact kinematics and the brain strain predicted by the Imperial College FE brain model. She compared deep neural networks with a lightweight XGBoost model designed specifically for rapid prediction; by reducing the input to a small number of features derived from rotational head motion, the XGBoost model retains useful predictive accuracy while running in under a tenth of a second, a fraction of the computation time of a full finite element simulation.

A workable pipeline for brain-health surveillance in contact sport: head-impact kinematics from instrumented mouthguards feed directly into rapid, whole-brain and regional strain predictions, with active collaborations across rugby, boxing, and other contact sports. The model has been packaged into an open-source desktop application that lets users load head-impact kinematics, generate strain predictions in batch, and export the results for further analysis, a practical bridge between wearable sensors and computational brain biomechanics. The long-term goal is to make brain deformation itself, not just head motion, part of near real-time brain-health monitoring in sport.