Variabilités émotionnelle et neurophysiologique dans la perception olfactive: étude par l’intelligence artificielle

The subjective and emotional dimension of odors is a main component of olfactory perception. It allows the creation of approach or avoidance behaviors essential to the survival of the individual. Indeed, it participates in the prevention of potential dangers, in nutrition and in social communication. The way in which these olfactory affects are manifested in the nervous system is still not fully understood today. Moreover, the high inter-individual variability in olfactory perception no longer needs to be demonstrated and yet it remains little explored at the present time. In fact, due to the lack of adequate analysis tools, this variability is not taken into account in studies concerning the neurophysiology of emotion in olfactory perception. Thus, during this thesis project, we sought to remove these methodological barriers with approaches from artificial intelligence. Indeed, we used data science methods in order to understand how emotional diversity in odors is reflected at the neurophysiological level. We were specifically interested in exploring exceptional patterns and subgroups discovery. These descriptive methods allow us to extract precise information about the common behavior of a group of individuals, but also to give us insights at the individual level. Thus, we applied them to different neurophysiological measures of emotion: autonomic nervous system measures and central nervous system measures (fMRI and EEG) through three experimental studies. In the first study, the hedonic character (pleasant/unpleasant) and intensity of the odor were described by specific physiological reactions. In the second study, the pleasant and unpleasant components of the odor were localized spatially in the piriform cortex and amygdala. In the third study, the temporal dimension was explored in the frontal cortex. The use of these data mining approaches is particularly suitable because the descriptive patterns that emerge are consistent with past studies. Indeed, study 1 supports the hypothesis of a stronger somatic response to aversive (unpleasant and intense) olfactory stimuli. Study 2 is in agreement with studies showing asymmetric processing of unpleasant odors in the anterior piriform cortex. Study 3 shows a delayed response for pleasant odors and an instantaneous response for unpleasant ones already demonstrated before. Moreover, these approaches allow us to explore the individual diversity present for each description found. The study of variability indicates a greater heterogeneity for pleasant odors than for unpleasant odors concerning the activity of the peripheral nervous system (study 1) and the cerebral localization (study 2). This trend is reversed when considering the temporal dimension. Indeed, pleasant odors create a very precise response (late bilateral activity in the low frequencies) whereas different responses are observed according to the individuals for unpleasant odors. In order to allow non-specialists in the field of data science to apply this type of approach to their own data, we have suggested, in a fourth study, a methodological protocol. This tutorial describes step by step the application of subgroups discovery on skin conductance data. We also carried out a theoretical work in a fifth study. This one offers a reflection on the analysis of emotions in real conditions outside the laboratory. It suggests the use of data science methods as a possible solution to analyze the numerous and heterogeneous resulting data.

Thesis