< Symposia

Symposium Session 2 - Mapping Emotions in the Brain Beyond Localization: How Neuroimaging and Machine Learning Can Reshape Contemporary Theoretical Frameworks

Chair: Patrik Vuilleumier¹; ¹University of Geneva (Switzerland)
Presenters: Kevin LaBar, Patrik Vuilleumier, Heini Saarimäki, Ajay Satpute

Emotions are a central ingredient of the mind and behavior but continue to generate intense debate across psychology and neuroscience. Different theoretical frameworks are generally opposed to describe their nature, origin (innate vs. learned), adaptive role, as well as functional organization (modular or continuous). Discrete emotion theories propose a limited set of biologically hardwired emotions such as fear, anger, or joy, each tied to specific neural circuits, whereas dimensional theories conceptualize emotions as gradients along just a few dimensions like valence and arousal, and componential theories regard them as resulting from an interaction of multiple processes including cognitive evaluations (appraisals), action tendencies, and physiology changes. Going further, constructivist models argue that emotions are not hardwired categories but emergent states shaped by bodily signals, cognition, language, and culture. This symposium will review how recent neuroscience approaches may shed new light on these questions through a combination of novel paradigms and novel methodologies, particularly functional neuroimaging combined with machine learning, computational modelling, and artificial intelligence tools. Speakers will illustrate whether and how recent brain research supports, challenges, or instead extends these classic theoretical models. Among common themes, the talks will showcase that, regardless of theoretical perspective, neuroscience results point to multiple, widely distributed, and partly overlapping brain systems engaged during various emotional experiences. Our survey of the field should help to surpass a strict opposition between past models and go some way toward highlighting elements of convergence or complementarity rather than absolute divergence.

Presentations

The Semantic Space Organization of Fifteen Emotional States Decoded from fMRI Data

Kevin LaBar¹; ¹Duke University (USA)

Theoretical models emphasize that categorical factors, dimensional factors, or their combination may define the semantic space organization of emotion representations. While recent behavioral work has applied innovative multivariate methods for testing these theories, neuroscientific assessments remain limited due to a focus on a small number of emotions, single theoretical perspectives, univariate methods, or small sample sizes. We overcame these limitations in a comprehensive functional neuroimaging study where participants (N=136) viewed 150 normed movie clips that reliably induced 15 different emotional states spanning positive, negative, and neutral valence. Representational similarity analysis yielded a robust correspondence between brain and behavioral responses. Hierarchical clustering of the representation dissimilarity matrices revealed several meaningful clusters of emotions such as fear/anxiety, and joy/amusement. Partial least squares discriminant analysis achieved strong decoding performance from whole-brain fMRI data that successfully predicted each of the 15 emotion categories using subject-independent cross-validation with high AUC scores. Importance maps showed that these emotion representations spanned cortical, limbic, and subcortical regions and were not restricted to one (or more) of the canonical large-scale intrinsic brain networks. Informational content analysis and a Bayesian model comparison supported the categorical nature of the emotion representations relative to an arousal-valence circumplex model, with maximum classification accuracy requiring, at minimum, a 7-dimensional space projection. Although not favored in the classification of discrete emotional states, arousal and valence representations were successfully decoded from the whole-brain data using multivariate regression models. These findings provide novel insights into the theoretically-motivated semantic architecture of emotion representations in the human brain.

Brain Network Dynamics and Functional Components of Emotion

Patrik Vuilleumier¹; ¹University of Geneva (Switzerland)

Emotions consist of adaptive responses to particular events based on their perceived value and trigger multiple changes in perception, memory, or action. By using fMRI in emotion eliciting paradigms with movie watching or interactive video gaming, our work shows that emotional events influence brain activity and cognitive functions not only during but also subsequent to transient affective responses. We find that distributed brain networks activate in parallel and contribute jointly, though to different degrees, to the generation of different emotions. These networks may not only encode the affective valence of events, along two different systems seemingly engaged by liking and wanting dimensions, but also other cognitive dimensions such as novelty, goal, or social significance, in line with a theoretical framework where emotion experience relies on several functional components driven by distinct appraisals. We also find that transient synchronization of these networks during emotional episodes engages a set of key brain structures in medial prefrontal cortex, insula, and basal ganglia, respectively involved in self-relevance, interoception, and motor programming. Finally, emotional episodes lead to prolonged changes in brain state after the eliciting events, with predominant impact on midline brain areas as well as other areas implicated in attention, memory, or executive control. Taken together, our findings support an account of emotions as dynamic changes in brain state emerging from embodied and action-oriented processes, which govern adaptive responses to the environment with both short-term and more sustained effects on behavior.

Mapping Emotional and Interoceptive Experiences in the Human Brain

Heini Saarimäki¹; ¹Tampere University (Finland)

Emotional experience refers to what is often called a subjective feeling: the discernment and description of our own internal state. Emotional experiences result from automatic changes in distinct domains such as interoception, action tendencies, motivation, and cognition. In a series of studies, we directly compared self-reports of emotional and interoceptive experiences with functional brain imaging and multivariate data analyses. Our aim was to characterize domain-specific changes, especially interoception, in relation to different emotions and different neural activity patterns. We induced emotional and bodily experiences during functional magnetic resonance imaging (fMRI) using four different tasks: autobiographical imagery, emotional movies, emotional stories, and interoceptive imagery. We collected experienced emotions and interoceptive changes using self-reports. All emotional tasks activated overlapping brain regions, including the anterior cingulate cortex, frontal pole, and middle and inferior temporal gyri. Interoceptive imagery also engaged brain regions partly overlapping with emotion-related activity, particularly in cortical midline, motor areas, insula, subcortical structures, and prefrontal cortex. Self-reported interoceptive experiences were associated with activation patterns in the secondary somatosensory cortex. However, this was not observed in the insula or primary somatosensory cortex. Overall, our results demonstrate a shared neural basis for emotions and interoception, as well as a link between specific interoceptive and emotional experiences, in keeping with theoretical models where interoceptive signals constitute a key ingredient of emotional experience.

Circular Reasoning in the Neuroscience of Emotion: Why Debates on Emotion Don’t Seem to End

Ajay Satpute¹; ¹University of California Los Angeles (USA)

For decades, studies have sought to map affective experiences to specific brain regions, but the picture remains unclear. In this talk, I explore why, showing how assumptions in study design and analysis (including machine learning approaches) - can shape results in surprising ways, leading to circular reasoning. Using fMRI data on fear, I reveal how the same data can tell entirely different stories, depending on the analytical approach. I use computational approaches to evaluate which theoretical models are best supported by the data in relation to theories that assume prototype (e.g. basic emotions) or heterogeneous (e.g., constructionist) representations of emotion. While my talk focuses on affective neuroscience, the theoretical ideas and analytical tools are also more broadly applicable for addressing both fundamental and translational questions in psychology and neuroscience more broadly.