AI-powered tool aids in traumatic brain injury investigations

The findings, published in Communications Engineering, introduce an advanced physics-based, machine learning framework that could revolutionise the way TBI cases are approached in forensic investigations.

TBI remains a significant public health concern, with potentially devastating and long-lasting neurological effects. For law enforcement and legal teams, establishing whether an impact caused an injury is essential, but the field has lacked a standardised, quantifiable method for such determinations. This new study demonstrates how AI, powered by mechanistic simulations, can offer data-driven predictions to help forensic experts and police accurately assess TBI outcomes in the context of reported assault incidents.

The AI framework, trained on anonymised police and forensic data, showed impressive predictive accuracy for a range of TBI-related injuries:

• 94% accuracy for skull fractures
• 79% accuracy for loss of consciousness
• 79% accuracy for intracranial haemorrhage (bleeding within the skull)

Each prediction was marked by both high specificity and sensitivity, ensuring low rates of false positives and false negatives.

“This research represents a significant step forward in forensic biomechanics. By leveraging AI and physics-based simulations, we can provide law enforcement with an unprecedented tool to assess TBIs objectively,” said lead researcher Antoine Jérusalem, Professor of Mechanical Engineering at the University of Oxford.

The framework is underpinned by a comprehensive computational model simulating the mechanics of the head and neck. It predicts how various impacts – such as punches, slaps, or blows to a flat surface – affect different regions of the skull. This allows an initial estimation of whether an impact could result in tissue deformation or stress. However, it is the upper AI layer that refines the predictions, incorporating additional factors such as the victim’s age, height, and other relevant metadata to predict the likelihood of specific injuries.

For training, the researchers used 53 anonymised police reports from assault cases. These reports provided a wealth of data, including factors such as the victim’s body type and the severity of the impact. This helped the model combine mechanical biophysical data with real-world forensic details, enabling accurate predictions of various injury outcomes.

The study revealed that the model's predictions were closely aligned with medical findings. For example, the most important factor in predicting skull fractures was the amount of stress experienced by the skull and scalp during the impact. Similarly, stress metrics related to the brainstem proved to be the strongest predictor of loss of consciousness.

Ms Sonya Baylis, Senior Manager at the National Crime Agency, highlighted the importance of this innovation: “Understanding brain injuries using innovative technology to support a police investigation, previously reliant on limited information, will greatly enhance the interpretation required from a medical perspective to support prosecutions.”

Despite its promising capabilities, the researchers emphasised that the AI model was not designed to replace the expertise of forensic and clinical professionals. Instead, the tool aims to provide an objective estimate of the likelihood that a reported injury was caused by a specific assault. It could also help identify high-risk situations, improve risk assessments, and aid in the development of preventive strategies to minimise head injuries.

“We can’t identify with certainty who caused an injury, but we can provide a probability based on the information available,” explained Professor Jérusalem. “The quality of the output is directly linked to the quality of the information fed into the model, which is why detailed witness statements remain vital.”

Dr Michael Jones, Researcher at Cardiff University and Forensics Consultant, added, “One of the challenges in forensic medicine is assessing whether the force involved in an injury aligns with the observed damage. With machine learning, every case contributes to refining our understanding of how injuries occur, from cause to outcome.”

The study, titled A Mechanics-Informed Machine Learning Framework for Traumatic Brain Injury Prediction in Police and Forensic Investigations, was a collaborative effort involving the University of Oxford, Thames Valley Police, the National Crime Agency, Cardiff University, Lurtis Ltd., and the John Radcliffe Hospital, among other partners.