NIH Research Festival
Employing a temporal and spatio-temporal classification approaches, we attempted to identify the prefrontal hemodynamic biomarkers that distinguish TBI and healthy subjects. To achieve this goal, first the hemodynamic response from a group of 34 healthy and 33 chronic TBI subjects while performing a complexity task is captured. We presented a novel preprocessing approach in identifying the fNIRS hemodynamic signals that encompass task-related activity by imposing certain restriction on the HbO and HbR signals followed by an extensive feature extraction/selection procedure. For the 12 extracted hemodynamics features, 4095 classification experiments were run to identify the optimum set of functional biomarkers through the wrapper feature subset selection method. This method utilizes machine learning classification algorithm as a black box to score different subsets of the hemodynamic features according to their predictive power. Finally, the largest possible accuracy in in characterizing the TBI subjects using the proposed method is determined by employing different classification algorithm for the selected set of features. The sensitivity of 84% was obtained for the selected feature set suggesting major contribution of the identified features in characterizing the differences between the TBI and healthy subjects.
Scientific Focus Area: Neuroscience
This page was last updated on Friday, March 26, 2021