Concurrent Prediction of Performance-Critical Cognitive States from Physiological Signals

Crew behavioral health is critical to spaceflight mission success. As future missions move further from Earth, crews will have to rely less on ground control for support. In its place, novel methods to monitor, understand, and respond to changes in their behavioral health could inform the timing of full cognitive assessments or automated crew-resource management decisions. Physiological signals offer a non-disruptive, objective method for capturing continuous information about crew cognitive state; however, limited work has evaluated predictive performance from combinations of these signals. In this work, we collected a multimodal suite of neurophysiological, psychophysiological, and behavioral signals from 31 participants (16F) while they completed the Multi-Attribute Task Battery II. Additionally, we collected gold-standard or proxy measures of cognitive states, including workload, attentional allocation, working memory, vigilance, and engagement. We present predictive models of these measures along a continuum and validate model performance on unseen data. For instance, we predict overall weighted NASA TLX scores on a continuous 100-point scale with mean absolute error (MAE) of 11.22 and with a Q2 of 0.47. To evaluate baseline model performance, we shuffle our TLX labels to align them with unrelated physiology, and we show that MAE increases (worsens) to 13.95 and Q2 decreases (worsens) to -0.31. We present predictive features and their coefficients, including respiratory tidal volume, oxygenated prefrontal hemoglobin, and deoxygenated prefrontal hemoglobin. Model type and performance metric necessarily change based on dependent variable, and we discuss the implications of these different strategies and performance metrics. Overall results show that physiological signals could provide critical insight into a crew’s cognitive state where other methods may prove impractical or infeasible.