Genetic Fuzzy Methodology to Predict Time to Return to Play from Sports-Related Concussion

Abstract

Sports related concussion (SRC) is a common health concern for young athletes. However, SRC is primarily managed from subjective assessment of symptomology and the unique response to concussive head impact, warrants more objective and personalized prognosis and management particularly related to the difficult decision of when to return to play. A data based objective approach is needed in this area, particularly to inform return-to-play decision making. We present a genetic fuzzy based methodology for predicting the time a player would need to return to play following a concussion. The predictions are based on connectivity data collected using Diffusion Tensor Imaging (DTI) data collected among a large cohort of athletes post SRC. Principal Component Analysis (PCA) is done for variable reduction. Fuzzy Bolt^© is used to train the GFS for regression and classification. For the regression case, the GFS model is trained to predict the return to play time needed whereas for classification, another GFS model is trained to predict if the return to play would be short ($\le 13$ days) or long ($> 13$ days). The dataset used for training included 34 datapoints, and the performance of the GFS was compared against neural networks and support vector machines. The paper presents the improvement obtained using GFS over neural networks and support vector machines.