Poster No:
2021
Submission Type:
Abstract Submission
Authors:
Kwangsun Yoo1,2,3, Young Hye Kwon4, Marvin Chun3,5
Institutions:
1Sungkyunkwan University, Seoul, Korea, Republic of, 2Samsung Medical Center, Seoul, Korea, Republic of, 3Yale University, New Haven, CT, 4Feinberg School of Medicine, Northwestern University, Chicago, IL, 5Yale School of Medicine, New Haven, CT
First Author:
Kwangsun Yoo
Sungkyunkwan University|Samsung Medical Center|Yale University
Seoul, Korea, Republic of|Seoul, Korea, Republic of|New Haven, CT
Co-Author(s):
Young Hye Kwon
Feinberg School of Medicine, Northwestern University
Chicago, IL
Marvin Chun
Yale University|Yale School of Medicine
New Haven, CT|New Haven, CT
Introduction:
Our recent innovation, connectome-to-connectome (C2C) state transformation modeling, can predict how an individual's functional connectome reconfigures to meet different cognitive contexts, such as behavioral tasks involving memory, emotion, and attention, solely using resting-state fMRI data [1, 2]. In this study, we expand the untapped potential of this versatile modeling framework to predict how the connectome subtly reorganizes in response to different visual scenes within the same movie, despite their similar and neutral cognitive content.
Methods:
We used the rest and movie-watching fMRI dataset (n=92) originally collected in Yoo et al. [1]. We excluded 15 individuals with only single session data (out of two visits) or less than 30 TR volumes for any of the six movie segments after censoring, resulting in a final set of n=77 for analysis. Image acquisition, (pre)processing, and experimental design are described in Yoo et al. [1].
In this study, we adapted the C2C transformation framework [1, 2] to construct models predicting an individual's connectome-wise reorganization from resting state to movie-watching states induced by separable visual scenes while viewing the abstract video, "Inscape." The C2C transformation involves a three-step process: 1) extraction of subsystems from the resting-state whole-brain connectome, 2) transformation of these extracted subsystems to the movie-watching states, and 3) construction of whole-brain connectomes for the movie-watching states based on the transformed subsystems. "Inscape" can be divided into six segments with distinct visual scenes [3], and we constructed and validated six C2C transformation models using an 11-fold cross-validation (CV) approach (70 training and 7 testing samples) iterated 100 times. Details of the C2C modeling framework can be found in the previous publications [1, 2].
We evaluated the models' connectome prediction in two ways. First, we assessed the specificity for movie segments. For each testing sample, we performed a paired t-test to compare spatial similarities between predicted and observed connectomes and used fingerprinting analysis [4] to estimate the identification rate of an individual's predicted connectome across the six movie segments. Second, we assessed whether connectome prediction preserves individual uniqueness. For each movie segment, we calculated the identification rate of an individual's predicted connectome across the seven testing samples.
Results:
The observed connectomes were most similar to the predicted connectomes of the corresponding segment (paired t-test ps<0.001) for all but one segment. The identification rate averaged across participants, CV folds, and iterations was 36.5% which significantly exceeds the chance level of 16.7% (p<0.001). In individual fingerprinting, the identification rate averaged across movie segments, CV folds, and iterations was 59.4% which significantly surpasses the chance level of 14.3% (p<0.001).
Conclusions:
Our study represents the successful application of the C2C transformation framework to predict whole-brain connectome-wise subtle reorganization in response to distinct visual themes within the same (or at least similar) cognitive context. The success of our model predictions suggests that even simple linear models can capture distinct connectome-wise reorganizations in response to visually similar stimuli, paving the way for deeper investigations into the dynamics of cognitive states.
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
fMRI Connectivity and Network Modeling
Multivariate Approaches
Task-Independent and Resting-State Analysis 1
Keywords:
FUNCTIONAL MRI
Modeling
Multivariate
1|2Indicates the priority used for review
Provide references using author date format
[1] Yoo, K. et al. (2022), ‘A cognitive state transformation model for task-general and task-specific subsystems of the brain connectome’, NeuroImage, vol. 257, pp. 119279.
[2] Yoo, K. et al. (2022), ‘A brain-based general measure of attention’, Nature Human Behaviour, vol. 6, no. 6, pp. 782-795.
[3] Vanderwal et al. (2015), ‘Inscapes: A movie paradigm to improve compliance in functional magnetic resonance imaging’, NeuroImage, vol. 122, pp. 222-232.
[4] Finn, E.S. et al. (2015), ‘Functional connectome fingerprinting: identifying individuals using patterns of brain connectivity’, Nature Neuroscience, vol. 18, no. 11, pp.1664-1671.