Neural Signatures of Ongoing Thoughts During Rest
Poster No:
957
Submission Type:
Abstract Submission
Authors:
Jin Ke1, Xiaochen Ding1, Taylor Chamberlain2, Anna Corriveau1, Hayoung Song1, Ziwei Zhang1, Taysha Martinez1, Laura Sams1, Monica Rosenberg1
Institutions:
1The University of Chicago, Chicago, IL, 2Columbia University, New York, NY
First Author:
Co-Author(s):
Introduction:
While we rest, our mind often spontaneously wanders from ourselves to others, from the past to the future. These thoughts can be elicited by precipitating events and also emerge in the absence of direct external stimuli. Self-generated thoughts and feelings have been suggested to reflect personal traits[1] and provide behavioral markers for mental disorders[2]. Building on an emerging literature that utilizes neuroimaging[3] and natural language processing[4] to study the neural correlates and content of self-generated thoughts, here, we introduce an annotated resting-state fMRI paradigm for nuanced understandings of associations between unconstrained thoughts and brain dynamics. We use functional brain connectivity patterns[5] to track dimensions, topics, and linguistic sentiment of spontaneous thoughts.
Methods:
Across two fMRI scan sessions, participants (N=50) completed four 10-min runs (32 rest periods total) of an annotated rest task in which they rested for 30s, verbally reported their ongoing thoughts for 10s, and rated their thoughts on 9 dimensions[6] using a slider bar (e.g., thinking about the future vs past). 5 thought topics (e.g., positive social memory) were extracted from the 9 dimensions using principal component analysis. We used a roBERTa-based model[7], a pre-trained sentiment analysis model based on Twitter posts, to conduct sentiment analysis on the recording transcriptions to generate probabilities of the speech being negative or positive.
A whole-brain functional connectivity (FC) pattern was generated for each 30s rest period using 268 functionally-defined ROIs[8]. We built support vector regression models with a leave-one-subject-out cross-validation approach to predict thought dimensions or topics (PCs) from FCs observed during the intermittent rest periods. These connectome-based models (CPMs)[9] were trained using data concatenated across all training subjects' rest periods and tested on each of the held-out subject's rest periods separately. Model performance was assessed by correlating predicted and observed values within subjects and comparing the mean within-subject correlation to a null distribution generated with permutation testing.
A whole-brain functional connectivity (FC) pattern was generated for each 30s rest period using 268 functionally-defined ROIs[8]. We built support vector regression models with a leave-one-subject-out cross-validation approach to predict thought dimensions or topics (PCs) from FCs observed during the intermittent rest periods. These connectome-based models (CPMs)[9] were trained using data concatenated across all training subjects' rest periods and tested on each of the held-out subject's rest periods separately. Model performance was assessed by correlating predicted and observed values within subjects and comparing the mean within-subject correlation to a null distribution generated with permutation testing.
Results:
Dimensions of ongoing thoughts correlated with each other as well as with speech sentiment decoded by linguistic analysis (Fig.1A). CPMs yielded above-chance predictive accuracy in predicting 5 out of 9 dimensions (Fig.2) and 4 out of 5 thought topics as well as positive (r=.047,p=.011) and negative (r=.049,p=.009) speech sentiment. Additionally, CPMs trained on positivity ratings predicted positive (r=.041,p=.030) and negative speech (r=-.106,p<.001) sentiment. Pairs of dimensions share significantly more overlapping edges than chance (Fig.1B). The number of shared edges predicts the absolute value of behavioral correlation between pairs of dimensions (Spearman's r=.591,p<.001). Further, whole-brain FC pattern similarity predicted thought similarity, such that rest periods with more similar FC patterns were accompanied by more similar self-reported ratings (mean within-subject correlation r= .089, p<.001).
·Figure 1. A. Behavioral similarity between pairs of thought dimensions and speech sentiment. B. Number of overlapping FCs between pairs of thought dimensions and speech sentiment.
·Figure 2. Functional connectivity during intermittent rest predicts dimensions of ongoing thoughts.
Conclusions:
Functional brain correlations between regions observed during intermittent rest encode the self-reported dimensions and topics as well as linguistic sentiment of ongoing thoughts. FC networks predicting these dimensions overlap with each other, suggesting related dimensions of ongoing thoughts share underlying neural correlates. Further, thoughts arise in the brain as topics that are decodable from brain activity patterns. Additionally, sentiment analysis provides complementary tools to decode the emotional components of ongoing thoughts in a linguistic space beyond self-reported ratings.
Emotion, Motivation and Social Neuroscience:
Emotion and Motivation Other 2
Higher Cognitive Functions:
Higher Cognitive Functions Other 1
Modeling and Analysis Methods:
fMRI Connectivity and Network Modeling
Keywords:
Cognition
Computational Neuroscience
Emotions
FUNCTIONAL MRI
Language
Other - Functional Connectivity
1|2Indicates the priority used for review
Provide references using author date format
[2] Smallwood, J. (2007). Mind-wandering and dysphoria. Cognition and Emotion, 21(4), 816-842.
[3] Karapanagiotidis, T. (2020). The psychological correlates of distinct neural states occurring during wakeful rest. Scientific reports, 10(1), 21121.
[4] Li, H. (2021). Exploring self-generated thoughts in a resting state with natural language processing. Behavior Research Methods, 1-19.
[5] Gonzalez-Castillo, J. (2015). Tracking ongoing cognition in individuals using brief, whole-brain functional connectivity patterns. Proceedings of the National Academy of Sciences, 112(28), 8762-8767.
[6] Ho, N. S (2020). Facing up to the wandering mind: Patterns of off-task laboratory thought are associated with stronger neural recruitment of right fusiform cortex while processing facial stimuli. Neuroimage, 214, 116765.
[7] Liu, Y. (2019). Roberta: A robustly optimized bert pretraining approach. arXiv preprint arXiv:1907.11692.
[8] Shen, X. (2013). Groupwise whole-brain parcellation from resting-state fMRI data for network node identification. Neuroimage, 82, 403-415.
[9] Shen, X. (2017). Using connectome-based predictive modeling to predict individual behavior from brain connectivity. nature protocols, 12(3), 506-518.