• affective computing
• biomedical signal processing
• bipolar patients
• convex optimisation
• electrodermal activity
• emotions

Electrodermal activity (EDA) can be considered as one of the most common observation channel of the autonomic nervous system (ANS) activity, and manifests itself as a change in electrical properties of the skin. Several previous works pointed out how EDA can be a powerful biomedical signal, although many issues arise while extracting information to infer on the subject’s psychophysiological state. This thesis aims at investigating how EDA can be a source of reliable and effective markers for the assessment of emotional responses in healthy subjects, and for the recognition of pathological mood states in bipolar patients. Throughout this dissertation, in-depth methodological and applicative studies involving EDA are described, including a critical review on the current state-of-the-art. As Continuous Deconvolution Analysis (CDA) has been recognized as one of the mostly used methods for EDA analysis, I first applied this model to distinguish different affective states in healthy volunteers. Emotions were elicited using standardized sets of pictures, sounds, caresses, and smells. Valence and arousal levels of such emotions were identified as the principal dimensions of affective responses. Results on this regard were consistent with the hypothesis that it is possible to objectively study ANS dynamics involved in the emotional processing, by properly analyzing the EDA.
In order to improve the statistical power of the EDA-derived features during multi-sensory emotional regulation, and to avoid some of the heuristic solutions and post processing steps of the conventional approach, I developed a novel computational model for the EDA analysis based on convex optimization methods. This model, called cvxEDA, describes EDA as a sum of three terms: the phasic component, the tonic component, and an additive white Gaussian noise term incorporating prediction errors, as well as measurement errors and artifacts. CvxEDA is physiologically inspired and fully explains EDA through a rigorous methodology based on Bayesian statistics, mathematical convex optimization, and sparsity. The algorithm was evaluated in three different experimental sessions to test its robustness to noise, its ability to separate and identify stimulus inputs, and its capability of properly describing the ANS activity in response to strong affective stimulation. Building on my previous CDA-based experimental results, outcomes of cvxEDA demonstrate higher accuracy than CDA in discerning elicited emotional states in healthy subjects. Within this methodological framework, concerning clinical applications involving psychiatric patients, I demonstrated that EDA strongly changed according to different mood states. This allows using EDA phasic and tonic components as suitable markers for discriminating pathological mood states in bipolar patients.