A method combining finite element simulation and machine learning for performing lower extremity firearm injury diagnosis

As global conflicts escalate, addressing potential human injuries on the battlefield becomes increasingly critical. However, ethical considerations preclude the use of human experimentation in injury research. Fortunately, advancements in computer simulation, particularly through numerical methods and machine learning algorithms, offer promising avenues for understanding mechanisms of injury without resorting to unethical practices. In this study, we explore the integration of the finite element analysis method with machine learning algorithms to expedite the diagnosis of human lower extremity firearm injuries. Initially,we will establish a finite element model of the human lower limb bones based on CT images and validated material properties of the fibula and tibia. Subsequently, we will simulate various combat scenarios by altering different loading conditions. By scripting parameterized simulations, we will generate a large dataset of simulation results. Upon obtaining finite element results for each simulated case, we employ supervised machine learning algorithms to classify wound characteristics, including entrance velocity, projectile type, and angle of incidence. Our findings demonstrate high diagnostic accuracies, with k nearest neighbor (KNN), multilayer perceptron(MLP),support vector machine (SVM) and random forest classifiers achieving accuracies of 85.5%, 86.5%, 86.5% and 86%, respectively.