Analysis of neural substrates and physiological responses involved in the processing of primary emotions

Résumé

Emotions are a key process in the evolution of species and in particular in the development of human beings. They influence most of the neuronal processes that take place in our day to day, from decision-making, communication and social relations, selective attention, learning and memory. Understand their physiology and neurobiology is a great challenge, still unanswered, which has been raised since Greco-Roman times. The large number of applications from the clinical, for the diagnosis and treatment of mood disorders, to the improvement of brain-computer interactions applicable in both patients and healthy individuals, is unimaginable; so that getting a system capable of recognizing emotions in real time is the holy grail of affective neuroscience today. However, the lack of a theoretical model that defines the term itself, as well as its primary components and mechanisms of action, causes a great variability between the results and computational models that try to solve the equation. In order to be able to establish a real-time emotion classification model, it is necessary to establish a series of previous parameters, such as the emotional model to follow, the optimal window time to extract the features that encode emotional information, the feature extraction method, how many and which are the features that represent the processing of emotional information and the appropriate algorithm to classify this type of information and signal. Based on the dimensional model of emotions, specifically the dimension of valence that characterizes the positive or negative degree of a stimulus generating responses of approach or withdrawal, respectively; we have tried to specify the parameters necessary to develop a computational model that allows us to recognize emotions on the scale of emotional valence focused on real-time applications. For this purpose, we have analyzed the electroencephalographic, electrocardiographic and skin temperature signals of 24 volunteers during emotional stimulation. This stimulation was carried out through an own-design audiovisual database that contained the same number of videos with content classified as positive and negative. The analysis of the data was developed taking into account two experimental approaches, one subject-dependent (SD) and another subject-independent (SI). The results obtained by SD approximation allowed us to elaborate a computational model based on the electroencephalographic (EEG) signal, achieving a precision of 0.989 (±0.013) according to f1-score. The model is based on a 12-second trial window, the non-linear method called wavelet packets for feature extraction, 20 pairs of frequency-location features distributed along much of the cerebral cortex and in the range of 8 to 45 Hz in the EEG spectrum, and the quadratic discriminant analysis and k-nearest neighbors classifiers. On the other hand, in the signal coming from the peripheral nervous system, specific response patterns were found for each emotional category, suggesting that the dimension of valence influences the response of the emotional somatic component. However, taking body response into account did not improve the accuracy of our computational model of recognition of positive and negative emotions; furthermore, the results found do not allow us, for the time being, to make inferences about the physiology of emotions. At the neurobiological level, patterns of inter-hemispheric asymmetry as well as rostro-caudal asymmetry suggest the existence of a neuronal circuit of emotional valence processing. However, in order to define this circuit precisely and to provide more evidence to the mechanism of action of the dimension of emotional valence, it would be necessary to continue with the studies of the relationships between the different highlighted areas and frequencies. The results obtained by the computational model based on the EEG signal proposed in this doctoral thesis, motivate the continuity of the study of emotions based on the dimensional model, with the objective of demonstrating the validity and reproducibility of the model proposed in real time, and in order to elaborate a theory that gathers the pathway and mechanisms of action of the dimension of emotional valence on both cerebral and corporal components.