Concussion classification via deep learning using whole-brain white matter fiber strains

Abstract

Developing an accurate and reliable injury predictor is central to the biomechanical studies of traumatic brain injury. State-of-the-art efforts continue to rely on empirical, scalar metrics based on kinematics or model-estimated tissue responses explicitly pre-defined in a specific brain region of interest. They could suffer from loss of information. A single training dataset has also been used to evaluate performance but without cross-validation. In this study, we developed a deep learning approach for concussion classification using implicit features of the entire voxel-wise white matter fiber strains. Using reconstructed American National Football League (NFL) injury cases, leave-one-out cross-validation was employed to objectively compare injury prediction performances against two baseline machine learning classifiers (support vector machine (SVM) and random forest (RF)) and four scalar metrics via univariate logistic regression (Brain Injury Criterion (BrIC), cumulative strain damage measure of the whole brain (CSDM-WB) and the corpus callosum (CSDM-CC), and peak fiber strain in the CC). Feature-based machine learning classifiers including deep learning, SVM, and RF consistently outperformed all scalar injury metrics across all performance categories (e.g., leave-one-out accuracy of 0.828–0.862 vs. 0.690–0.776, and .632+ error of 0.148–0.176 vs. 0.207–0.292). Further, deep learning achieved the best cross-validation accuracy, sensitivity, AUC, and .632+ error. These findings demonstrate the superior performances of deep learning in concussion prediction and suggest its promise for future applications in biomechanical investigations of traumatic brain injury.

Mesh Convergence Behavior and the Effect of Element Integration of a Human Head Injury Model

“…The advantage of a response vector over scalar metrics in injury prediction was supported by a recent study, where feature-based machine learning/deep learning utilizing all of the voxelwise WM fiber strains (vs. element-wise maximum principal strain evaluated here) significantly outperformed all scalar injury metrics via conventional logistic regression for concussion prediction. Comparable performances were also retained when using a subset of the response vector via feature-selection 5 .…”

Performance Evaluation of a Pre-computed Brain Response Atlas in Dummy Head Impacts

“…Most biomechanical studies have so far focused on identifying the “best” injury predictor to assess the probability of the occurrence of a binary injury 8,15,17,19,27,28,30,33,40,44 . There is a large gap of knowledge on how external head impacts, through induced brain responses at the time of impact, are related to subsequent, specific neurological disorder and injury severity often observed at a later time in the clinic.…”

Predicting Concussion Outcome by Integrating Finite Element Modeling and Network Analysis

“…LOOCV was selected to maximize our training set for validation given our small dataset size (n = 53) and for the low bias that LOOCV exhibits (Beleites et al, 2005). Other studies have employed LOOCV for the reconstructed football impacts and supplemented with out-of-bootstrap cross-validation to address LOOCV's sometimes high variance, but it did not qualitatively alter their findings (Beleites et al, 2005;Cai et al, 2018). We also reported sensitivity and specificity of the cross-validated predictions.…”