Schedule of Events | Symposia

Symposium Session 9 - Decoding spontaneous thought from neural activity

Symposium Session 9: Tuesday, April 1, 2025, 1:30 – 3:30 pm EDT, Grand Ballroom

Chairs: Aaron Kucyi¹, Julia Kam²; ¹Drexel University, ²University of Calgary
Presenters: Julia Kam, Aaron Kucyi, Matthias Mittner, Kalina Christoff

Understanding the neural basis of mind-wandering and spontaneous thought is an emerging topic in contemporary cognitive neuroscience. In an era dominated by experimental models in cognitive neuroscience that employ structured tasks, the study of spontaneous cognition faces a unique challenge: how can mental experience that is initiated without any experimental triggers—and that is unpredictable even by the experiencer themself—be measured and attributed to neural signals? In our symposium, speakers will present novel paradigms designed to tackle this challenge and demonstrate the decoding of spontaneous thought from neural activity with increasing precision. Our topics include novel applications of machine learning, computational modeling, and precision neuroimaging for the prediction of mind wandering and spontaneous thought from neuroimaging and electrophysiological data. Julia Kam will present evidence for a comprehensive set of brain-to-experience mapping of the phenomenological features of ongoing thoughts and the successful detection of their occurrence during naturalistic behavior using EEG. Aaron Kucyi will present an idiographic (person-specific) approach to predictive modeling of mind-wandering from resting-state fMRI data using precision neuroimaging. Matthias Mittner will present a Hidden Markov Model analysis of fMRI-pupillometry to detect switches between on-task and mind wandering states. Kalina Christoff will describe the differential role of subsystems of the default network in the spontaneity and automaticity in thought, as distinguished within the Dynamic Framework of Thought. Together, our symposium will highlight different experimental and analytic approaches that will set the foundations for a more comprehensive understanding of the neural basis of spontaneous thoughts moving forward.

Presentations

Predicting spontaneous thoughts using electroencephalogram during naturalistic behavior

Julia Kam¹; ¹University of Calgary

Humans engage in a continuous stream of ongoing mental experience. Although recent fMRI work revealed the functional connectivity patterns underlying several thought dimensions during experimental tasks, little is known about the electrophysiological basis of these thought dimensions in more naturalistic settings and the extent to which we can predict these ongoing thoughts using EEG and machine learning. To address this, we first examined the electrophysiological signatures of ongoing thoughts during naturalistic tasks in seven participants across seven recording sessions. We then combined deep learning algorithms with electrophysiological data to determine the utility of these signals in predicting thought dimensions. Based on a total of 49 datasets, our results revealed distinct oscillatory markers of seven dimensions of ongoing thought as participants completed any computer-based activities they wished to perform. In addition to identifying electrophysiological markers consistent with those observed in experimental settings for internally oriented and freely moving thoughts, we found novel patterns for off-task, goal-oriented, sticky, self and others-oriented thoughts. Importantly, applying deep learning algorithms on electrophysiological data reliably detected the occurrence of all seven thought dimensions based on different EEG features, highlighting the utility of combining deep learning approaches with EEG to detect covert, mental states. Together, these results assembled a comprehensive set of brain-to-experience mapping of the phenomenological features of ongoing thoughts and established the successful detection of their occurrence during naturalistic behavior. Our findings provide an important step towards predicting thought patterns in the real world with clinical implications for establishing biomarkers of atypical thought patterns.

Predictive neural modeling of resting-state spontaneous thought: an idiographic approach

Aaron Kucyi¹; ¹Drexel University

Spontaneous thoughts arise at unpredictable moments. There has been growing scientific interest in using machine learning to model and predict the momentary occurrence of spontaneous mental experiences from neural activity during task-free (resting) states. Commonly, training data include neural features derived from a population of individuals, and testing is performed on held-out data in one or more individuals. However, individual differences in the nature of spontaneous thought raise the critical question of whether population-derived neural models can generalize to individuals. We intensively-sampled three individuals who each reported hundreds of episodes of spontaneous off-task thought across multiple fMRI sessions while engaged in a simple resting state with intermittent experience sampling. Idiographic (i.e., person-specific) connectome-based predictive modeling revealed time-varying whole-brain functional connectivity patterns that consistently predicted momentary off-task thought within individuals but did not fully generalize across individuals. Predictive features were unique to each individual but commonly included the default mode network (DMN), the network most typically implicated in spontaneous thought. However, computational removal of the DMN had a minimal impact on prediction performance, and other networks such as sensorimotor were sufficient for prediction within all individuals. Predictive models of off-task thought and sustained attention from previously published fMRI studies largely failed when applied to intensively-sampled individuals, further highlighting the need for idiographic models. Our work offers strong evidence for person-specific neural representations of spontaneous thought with important implications for the interpretation and practice of resting-state fMRI. Our findings further underscore the broader value of idiographic approaches to predictive brain-experience relationships.

Modeling dynamical transitions between on- and off-focus states using Hidden-Markov Models

Matthias Mittner¹; ¹The Arctic University of Norway

Previous research has shown that large-scale brain networks such as default-mode network (DMN), executive network and dorsal-attention network are involved in mind-wandering (MW). However, current state-of-the-art methods for predicting MW based on neural data are lacking an explicit model for the temporal evolution and dynamic switches between on-task and MW states. Based on a theoretical model of mind-wandering that postulates that attentional shifts are modulated by norepinephrinergic (NE) activity, we investigate attentional switches using data from a combined fMRI and pupillometry study (N=27). This model assumes the existence of a latent off-focus state, characterized by transiently elevated tonic NE activity, which mediates transitions between on-task and MW. Because of the transient nature of this state, it is difficult to measure with standard methods. To overcome this problem, we implemented a modified Hidden-Markov model (HMM) fit to behavioural and pupillometric data that allows for the modeling of dynamic task transitions between latent states. Crucially, the model is also informed by self-reported MW in the form of thought-probes, and therefore provides interpretable latent states with specific signatures. By projecting the sequence of latent states extracted by the HMM into the fMRI data, we can extract specific brain signatures of on-task, MW and most importantly, the elusive off-focus state. We find that our analysis provides important insights into the dynamics of mind wandering and contributes to disentangling the enigmatic involvement of the DMN and its subnetworks in MW. Finally, we apply the model to a separate dataset to validate its generalizability and robustness.

Distinguishing between spontaneity and automaticity using the Dynamic Framework of Thought

Kalina Christoff¹; ¹University of British Columbia

Recent precision-fMRI findings reveal two distinct default mode networks (DMNs), DMN-A and DMN-B. How do these two DMNs relate to spontaneity and automaticity in thought, as distinguished within the Dynamic Framework of Thought (Christoff et al., 2016; Girn et al., 2020)? The newly identified DMN-A appears to be closely linked to episodic thought and scene construction, while DMN-B has been linked to semantic thought and social meaning-making. Our fMRI findings with highly experienced meditators suggest that the flexible interactions between subcortical and cortical regions may be the most distinctive feature of spontaneously arising thoughts. Thus, we have observed a spread of activation from subcortical DMN-A structures (e.g., medial temporal lobe) to cortical DMN regions, during the seconds preceding subjective reports of spontaneously arising thoughts in highly experienced meditators. Less experienced meditators, on the other hand, report fewer spontaneously arising thoughts and do not reliably recruit subcortical DMN-A regions prior to spontaneous thought reports, although they do show robust recruitment of cortical DMN regions. Thus, flexible interactions between subcortical and cortical regions of DMN-A appear to be a feature of spontaneous mental arisings. Less is known about the DMN-B. However, the involvement of cortical DMN-B regions in semantic processing and social meaning making hint towards a possible contribution towards the automaticity of thought through semantic constraints. Overall, the subcortical-to-cortical interactions within the DMN-A appear at present as the most promising target of investigation for distinguishing spontaneity from automaticity in thought, but DMN-B interactions remain an important topic for further investigation.