Identifying Neurophysiological Biomarkers Derived From fNIRS To Quantify Cognitive Workload
Monday, June 17, 2024
11:00 AM-1:00 PM
BIOMED Master's Thesis Defense
Title:
Identifying Neurophysiological Biomarkers Derived From fNIRS To Quantify Cognitive Workload During a UAS Mission
Speaker:
Cooper Molloy, Master's Candidate
School of Biomedical Engineering, Science and Health Systems
Drexel University
Advisor:
Kurtulus Izzetoglu, PhD
Associate Professor
School of Biomedical Engineering, Science and Health Systems
Drexel University
Details:
Safety critical tasks include missions where errors in execution can lead to negative consequences such as loss of life, financial loss, and environmental damage. Unmanned Aerial Systems (UAS) operators often perform such critical missions, and operator error has been shown to be a primary contributor to UAS mishaps. Cognitive workload, the mental effort required to complete a task, has been implicated as a factor that impairs UAS operator performance and increases mishaps. Thus, quantifying cognitive workload in UAS operators is crucial for performance enhancement and safety by enabling detection and management of cognitive overload and inadequate training. Traditional methods of measuring cognitive workload, such as the NASA-TLX and behavioral performance metrics, feature limitations that reduce the validity of their assessment of true cognitive workload. Neurophysiological measures, particularly functional Near Infrared Spectroscopy (fNIRS), have shown promise as direct, objective, non-intrusive tools to measure cognitive workload. Hence, the goal of this study is to evaluate the use of neurophysiological biomarkers derived from fNIRS measures in quantifying cognitive workload experienced during complex real-world UAS operations.
To assess the utility of fNIRS in quantifying cognitive workload, novice UAS operators (n=35) performed complex search and target find scenarios using a high-fidelity, commercially available UAS training simulator. Low- and high-workload conditions were induced by changing weather conditions. Neurophysiological, performance, and questionnaire data were collected for each scenario. Linear discriminant analysis, support vector machine, logistic regression machine learning classifiers were used to determine the classifier and feature set with the best predictive power. Leave one group out cross validation (LOGOCV) was implemented to generate test accuracies and F1-scores for each classifier. Forward and backward sequential feature selection were performed for both SVM and LR classifiers: all feature selected subsets contained fNIRS biomarkers, and all classifiers performed best with fNIRS biomarkers in the feature set. The LR classifier trained on backward selected features was shown to have the highest predictive power (test accuracy = 0.82 ± 0.00, F1-score = 0.87 ± 0.01). This study demonstrates the utility of neurophysiological measures in predicting cognitive workload.
Contact Information
Natalia Broz
njb33@drexel.edu