Combining fast and slow fMRI sampling rates can enhance predictive power in resting-state data

Area of Science:

• Neuroimaging
• Brain Dynamics
• Magnetic Resonance Imaging

Background:

• Functional magnetic resonance imaging (fMRI) technology is advancing, increasing spatio-temporal resolution for brain dynamics research.
• High sampling frequencies are assumed to yield more informative data and mitigate physiological noise, leading to the discard of lower temporal resolution datasets.
• Current MRI technology may be underutilized by exclusively collecting data at the fastest available rate.

Purpose of the Study:

• To investigate if collecting fMRI data at multiple timescales (multi-rate) can improve information gain about brain dynamics.
• To test the hypothesis that combining data from slow and fast repetition times (TR) yields greater insights than single-rate acquisition.
• To compare the predictive power of multi-rate fMRI datasets against single-rate datasets.

Main Methods:

• Analysis of a resting-state fMRI dataset from 10 subjects with simultaneous slow (2150 ms) and fast (100 ms) TR acquisitions.
• Comparison of predictive performance between multi-rate datasets and single-rate datasets (including manually undersampled data).
• Evaluation of gender prediction accuracy using composite features from multi-rate versus single-rate data.

Main Results:

• Combining fMRI data collected at slow and fast TRs demonstrated informative gains in predictive power.
• Single-rate dataset performance showed diminishing returns when data was manually undersampled.
• Multi-rate datasets achieved higher accuracy in gender prediction compared to single-rate datasets, highlighting gains in composite features.

Conclusions:

• Collecting fMRI data at multiple timescales, rather than solely at the fastest rate, can significantly enhance predictive power.
• Multi-rate fMRI acquisition aligns with theoretical predictions, showing that combined slow and fast sampling rates yield superior features in certain contexts.
• This study suggests a paradigm shift in fMRI data acquisition, advocating for multi-rate strategies to maximize information extraction and brain dynamics understanding.