Deep learning method to assess brain injury risk

Abstract

Model-based brain injury criteria can present higher potential to predict injury than global head kinematic parameters. Numerical head injury prediction tools are time consuming and require Finite Element (FE) skilled users. To address these difficulties, a deep learning technique was applied to an existing previously developed brain FE model for which an injury risk curve has been proposed in terms of maximum Von Mises stress to predict moderate diffuse axonal injury. A total of 4492 experimental head impacts coming from experimental helmet testing were considered as input for the analysis. Each input was expressed in terms of three linear accelerations and three angular velocities versus time, when the target metric was the time history Von Mises Stress () curve computed within the brain via the FE analysis. The architecture used for the Deep Learning (DL) model is the U-Net and four models based on it were evaluated. The dataset was split into three datasets dedicated for learning and testing. The quality of the DL models were assessed via the Maximum Absolute Error between FE and DL models computed brain maximum Further, a regression analysis of brain response with both methods was conducted. The results demonstrated that deep learning methods can be applied in the context of brain response estimation when helmet assessment is considered without any FE computation, only by considering the 6 D head kinematic vs time demonstrating that the deep learning approach should be further developed for research and industrial applications.