G-Induced Loss of Consciousness (GLOC) Predictive Model Development

Rinkel, Bridget

Etd

G-Induced Loss of Consciousness (GLOC) Predictive Model Development

Public Deposited

Military aircrew are often exposed to sustained +Gz in the high performance cockpit, which can reduce cerebral blood volume and tissue oxygenation, putting them at risk of G-Induced Loss of Consciousness (GLOC). GLOC can temporarily render aircrew unable to control their aircraft and is a major threat to aircrew safety and mission execution. Aircrew would benefit from having an early warning system that alerts them to impending GLOC based on data collected in operationally relevant environments. Data collected from the United States Air Force 711th Human Performance Wing’s GLOC study, including EEG, eye tracking, heart rate, and breathing rate, were used to develop machine learning-based models that provide accurate early indication of the onset of GLOC. Three model systems were designed and evaluated against each other to predict impending GLOC: 1) in unenrolled participants, whose data the model had not been trained on, 2) in enrolled participants, whose data the model had been trained on, and 3) in enrolled participants, whose data were used to train personalized models, to predict impending GLOC based on individual physiology. Accuracy, efficiency, and operational relevance were evaluated and optimized across models. Models were able to predict impending GLOC with 84.9% balanced accuracy up to 15 seconds before it occurred. The highest accuracies were obtained in System 2, by making predictions on “enrolled” participants. The lowest accuracy measures were found in System 1, by making predictions on “unenrolled” participants. The models were trained to predict impending GLOC based on physiological data collected in an operationally relevant environment and the tradeoff in model performance based on data collection and system objectives was demonstrated. The models help define the time course of development of GLOC from a physiologic and neurologic perspective and allow identification of features with the strongest predictive power for real-time risk assessment.

Creator