A statistical approach for segregating cognitive task stages from multivariate fMRI BOLD time series.

Area of Science:

• Neuroscience
• Cognitive Science
• Machine Learning

Background:

• Multivariate pattern analysis offers insights into human cognition and its disorders using neuroimaging.
• Functional magnetic resonance imaging (fMRI) blood oxygenation level dependent (BOLD) time series data contain rich information about brain activity.

Purpose of the Study:

• To develop and validate a novel methodological approach for distinguishing cognitive processing stages from fMRI BOLD time series.
• To apply this method to a virtual reality radial arm maze (RAM) task in healthy adults to study working memory and decision-making.

Main Methods:

• Utilized multivariate statistical/machine learning and time series analysis.
• Employed linear classifiers, multivariate test statistics, and time series bootstraps.
• Confirmed experimenter-defined task stages with an unsupervised Hidden Markov Model approach.

Main Results:

• Successfully discriminated between cognitive stages (encoding/retrieval, choice, reward, delay) in brain areas crucial for decision-making and working memory.
• Observed reduced discrimination during impaired behavioral performance in the dorsolateral prefrontal cortex (DLPFC), but not the primary visual cortex (V1).
• DLPFC differentiated memory load stages, while V1 differentiated visual-spatial aspects, highlighting region-specific cognitive roles.

Conclusions:

• The developed methodology effectively separates and attributes different cognitive information processing aspects to specific brain regions based on fMRI data.
• This approach provides a powerful tool for investigating the neural basis of complex cognitive tasks and their disruptions.