State descriptions drive predictions of individual cognitive traits from time-varying FC
Poster No:
1477
Submission Type:
Abstract Submission
Authors:
Christine Ahrends1, Mark Woolrich2, Diego Vidaurre3
Institutions:
1Aarhus University, Aarhus, Aarhus, 2University of Oxford, Oxford, Oxfordshire, 3Aarhus University, Aarhus, Aarhus C
First Author:
Co-Author(s):
Introduction:
Observing a person's behaviour over time is how we understand the individual's personality, cognitive traits, or psychiatric condition. The same should apply at the brain level, where we may be able to gain crucial insights by observing the patterns in which brain activity unfolds over time. Recent approaches allow predicting individual traits, such as age or cognitive traits, from time-varying or dynamic functional connectivity (FC). However, it is still unclear which aspects of this spatiotemporal level of description best characterise subject differences or predict individual traits. Time-varying or dynamic FC is represented using models of varying complexity (Lurie et al., 2019) that are estimated from modalities such as functional MRI or MEG. The parameters of these models describe different aspects of dynamic FC: In a state-based model, for instance, this may be the connectivity strength between nodes in the network of a transient FC state, or the transition probabilities between all states. By interrogating these parameters, we may be able to understand which aspects of dynamic FC are unique to an individual and related to their specific traits and which aspects are common between individuals.
Methods:
We here use resting-state fMRI data and behavioural measures of cognitive traits from 1,001 subjects from the Human Connectome Project (HCP) (van Essen et al., 2013). We use a Hidden Markov Model (HMM) to describe time-varying FC in the group of subjects (Vidaurre et al., 2017). We then estimate individual-level parameters of the HMM and construct from them the Fisher kernel (Jaakkola et al., 1999; Jaakkola & Haussler, 1998; Ahrends et al., 2023). The Fisher kernel is a mathematically principled approach for using a generative probabilistic model like the HMM in a prediction model or classifier. The HMM parameters can be categorised into two types: parameters related to the transitions between states and parameters related to the state descriptions (i.e., connectivity strength between brain regions). To test the influence of the different types of parameters, we construct three different kernels: one using all parameters, one using only transition parameters, one using only state descriptions. We then predict individual cognitive traits separately from each of these kernels and compare their prediction accuracies.
Results:
We found that state description parameters were the most relevant features for the Fisher kernel predictions in all cognitive traits. Both the full kernel and the kernel containing only state parameters predicted cognitive traits at high accuracy. The prediction accuracy was significantly reduced when state features were removed, while removing transition features had no significant effect. One reason for the dominance of state parameters may simply be that the state parameters outnumber the other parameters. However, reducing the number of state parameters to match the number of other parameters did not significantly affect accuracy. This indicates that the content of the state descriptions is more relevant for the prediction of cognitive traits than information about state transitions.
·Effects of removing sets of parameters from HMM Kernels
Conclusions:
Cognitive traits can be accurately predicted from time-varying FC using the HMM-Fisher kernel approach. We here showed that descriptions of transient FC states are highly predictive of individual cognitive traits, whereas state transitions carry little information about individuals. This suggests that differences between individuals in cognitive traits are linked to changes in connection strengths within resting-state functional networks.
Modeling and Analysis Methods:
Classification and Predictive Modeling 1
fMRI Connectivity and Network Modeling 2
Methods Development
Task-Independent and Resting-State Analysis
Keywords:
Cognition
Data analysis
Machine Learning
1|2Indicates the priority used for review
Provide references using author date format
Jaakkola, T. (1999), ‘Using the Fisher kernel method to detect remote protein homologies’, Proc Int Conf Intell Syst Mol Biol, pp. 149-158.
Jaakkola, T. (1998), ‘Exploiting Generative Models in Discriminative Classifiers’, NIPS, 11, pp. 487-493.
Lurie, D. (2019), ‘Questions and controversies in the study of time-varying functional connectivity in resting fMRI’, Netw. Neurosci., 4(1), pp. 30-69.
Vidaurre, D. (2017), ‘Brain network dynamics are hierarchically organized in time’, Proc Natl Acad Sci U S A, 114(48), pp. 12827-12832.
Van Essen, D. C. (2013), ‘The WU-Minn Human Connectome Project: an overview’, Neuroimage, 80, pp. 62-79.