Multivariate neural pattern changes reflect within-subject shifts in subjective interpretations
Poster No:
964
Submission Type:
Abstract Submission
Authors:
Clara Sava-Segal1, Tory Benson1, Emily Finn1
Institutions:
1Dartmouth College, Hanover, NH
First Author:
Co-Author(s):
Introduction:
Dynamically changing percepts can be reconstructed from multivariate neural representations[1]. This is usually studied with simple bistable stimuli in which people fluctuate between two mutually exclusive percepts. However, real-world ambiguities, like social contexts, tend to be more subjective, eliciting a wider range of spontaneous interpretations. Work with high-level, ambiguous stimuli (e.g., naturalistic narratives) shows that neural activity differs based on self-generated interpretations[2-4]. Thus, differences in neural activity are associated with differences in interpretations across subjects. However, it remains unknown if and where multivariate activity patterns change within subjects as individuals shift between interpretations of the same information. Given that we often encounter alternative explanations for the same event in social situations (e.g., a friend's opinion), we gave subjects an innovative, well-validated behavioral task that mimics ambiguous social scenarios. We tested the hypothesis that within-subject multivariate pattern shifts would predict self-reported shifts in subjective interpretations.
Methods:
During fMRI scanning (3T, voxel size=2.7mm3, TR=1.057), subjects(n=10) performed a task where, on each trial, they were presented with ambiguous photographs and interpreted them. Next, on experimental trials only, they read another possible interpretation sourced from a real subject (alternate interpretation). They then completed an appraisal task where they rated the likelihood of either just the self-generated interpretation (for control trials) or both the self-generated and the alternate interpretation (for experimental trials; Fig. 1). Scanning was conducted in an iterative design, where alternative interpretations were obtained from earlier subjects and selected in real-time, ensuring balanced distributions of 1)semantic distance between the self-generated and alternate interpretation, and 2)the identity of whom inputs were sourced from. Data collection is ongoing (target n=70).
Voxels were grouped into 100 cortical nodes using the Schaefer parcellation[5]. An event-related general linear model GLM) approach was applied. Each event was modeled by the onset and offset of a given trial phase (e.g., initial or second viewing;Fig. 1) convolved with the hemodynamic response function, yielding a beta weight for each trial phase per voxel. We computed neural shifts as the cosine distance (1-r) between multivoxel patterns in a given node between the two viewing phases of a trial (Fig. 2a). To predict the effect of condition, a linear mixed effects model was run per parcellation node predicting neural shifts by condition. Separate linear models were fit per subject. Within experimental trials, a separate linear mixed effects model was used to identify the effect of neural shifts on behavioral shifts in interpretation ("reappraisal"). Beta estimates from these models are plotted. An uncorrected threshold of p<.05 was applied.
Voxels were grouped into 100 cortical nodes using the Schaefer parcellation[5]. An event-related general linear model GLM) approach was applied. Each event was modeled by the onset and offset of a given trial phase (e.g., initial or second viewing;Fig. 1) convolved with the hemodynamic response function, yielding a beta weight for each trial phase per voxel. We computed neural shifts as the cosine distance (1-r) between multivoxel patterns in a given node between the two viewing phases of a trial (Fig. 2a). To predict the effect of condition, a linear mixed effects model was run per parcellation node predicting neural shifts by condition. Separate linear models were fit per subject. Within experimental trials, a separate linear mixed effects model was used to identify the effect of neural shifts on behavioral shifts in interpretation ("reappraisal"). Beta estimates from these models are plotted. An uncorrected threshold of p<.05 was applied.
·Fig. 1
Results:
Compared to the control condition, the degree of neural shift between viewings of the same sensory input in the experimental condition increased on a cortical gradient, with the largest differences across nodes in the frontoparietal network (FPN) and the default mode network (DMN; Fig. 2a). This cortical gradient holds at the individual-subject level (Fig. 2a, right). This indicates that simply being presented with an alternative interpretation is enough to drive changes in multivariate activity to the same visual stimulus. Further, we see within a subset of regions, including regions in the DMN and FPN, neural shifts predict behaviorally reported reappraisal, suggesting that these regions are involved in the process of reframing the visual input to change how it is ultimately perceived.
·Fig. 2
Conclusions:
Using a task with ambiguous social content, we identified that discernable multivariate pattern changes when processing the same sensory input can predict shifts in behavioral interpretation.
Emotion, Motivation and Social Neuroscience:
Social Cognition 2
Social Interaction
Higher Cognitive Functions:
Higher Cognitive Functions Other 1
Novel Imaging Acquisition Methods:
BOLD fMRI
Perception, Attention and Motor Behavior:
Perception: Visual
Keywords:
Cognition
Experimental Design
Multivariate
Other - Social neuroscience
1|2Indicates the priority used for review
Provide references using author date format
2. Nguyen, M., Vanderwal, T. & Hasson, U. Shared understanding of narratives is correlated with shared neural responses. NeuroImage 184, 161–170 (2019).
3. Finn, E. S., Corlett, P. R., Chen, G., Bandettini, P. A. & Constable, R. T. Trait paranoia shapes inter-subject synchrony in brain activity during an ambiguous social narrative. Nat Commun 9, 2043 (2018).
4. Sava-Segal, C., Richards, C., Leung, M. & Finn, E. S. Individual differences in neural event segmentation of continuous experiences. Cerebral Cortex bhad106 (2023) doi:10.1093/cercor/bhad106.
5. Schaefer, A. et al. Local-Global Parcellation of the Human Cerebral Cortex from Intrinsic Functional Connectivity MRI. Cereb Cortex 28, 3095–3114 (2018).