Tracking Neurophysiological Markers of Auditory Change Detection and Consciousness for Clinical Assessment in Coma

Abstract

The present dissertation expands the utility of EEG-based tools for the clinical assessment of coma. First, auditory mismatch negativity (MMN) responses were recorded in healthy controls and in a case series of comatose patients (over 12 and 24 hours, respectively). The results (Chapter 2) showed that the MMN elicited by deviant sounds, particularly for duration stimuli, is extremely robust in full conscious state over the course of several hours at both the group and single-subject levels. However, preliminary results in three comatose patients provide further evidence that the MMN is present but fluctuates in detectability in coma. These findings highlight that repeated assessments and proper stimuli selection are essential when assessing this ERP component as neurophysiological predictor of coma emergence. Then, a follow-up study (Chapter 3) demonstrated the feasibility of multivariate pattern analysis in our sample as an automatic tool able to discriminate accurately between single-trials responses at single-subject level, providing further evidence of changes in auditory discrimination over time in coma patients. Additionally, a phase-based measure of functional connectivity in response to auditory stimuli and resting state was computed for the first time at both the sensor (electrode) and source levels in a dying comatose patient. This report provided a ML procedure able to discriminate (with perfomance accuracies above 90%) single-trial functional connectivity elicited by deviant sounds between a comatose patient and healthy controls; and showed at least a period of increased synchronized activity during resting state before the end of life in the coma patient. Taken together, our findings showed that the EEG/ERP responses here studied, are highly transient in acute coma over hours and days, suggesting that repeated assessments are crucial for their objective detection and the methods of analyses should be sensitive enough to capture such changes. Overall, this work illustrates the utility of EEG along with machine learning to individualized neurophysiological assessment of comatose patients.