Jumping over baselines with new methods to predict activation maps from resting-state fMRI.

Area of Science:

• Neuroscience
• Cognitive Neuroscience
• Neuroimaging

Background:

• Cognitive functional magnetic resonance imaging (fMRI) research typically uses averaged data across subjects.
• Significant inter-subject variability exists, observable in resting-state fMRI (rsfMRI) spontaneous brain activity.
• Connectome fingerprinting aims to predict task activation from rsfMRI using machine learning.

Purpose of the Study:

• Evaluate the novelty and robustness of current methods predicting task activation from rsfMRI.
• Address the underperformance of existing methods against baseline models, especially for whole-cortex predictions.
• Introduce and validate a modified approach to improve prediction accuracy.

Main Methods:

• Reviewed existing literature on predicting task activation from rsfMRI.
• Compared published methods against robust baselines, including group averaging.
• Developed and implemented a modified prediction method utilizing a single-vertex approach instead of brain parcellations.
• Empirically characterized prediction performance across the cortex.
• Investigated the relationship between individual prediction scores and behavioral differences.

Main Results:

• Many existing methods underperform compared to simple group-averaging baselines for whole-cortex predictions.
• The proposed single-vertex approach significantly improves prediction performance.
• Identified specific cortical regions where prediction is more or less successful.
• Demonstrated that individual prediction scores can correlate with individual behavioral differences in the task.

Conclusions:

• Current connectome fingerprinting methods often lack robust validation against strong baselines.
• A single-vertex approach offers a substantial improvement over parcellation-based methods for predicting task activation from rsfMRI.
• Understanding regional prediction success is crucial for interpreting results.
• Individual differences in brain activity patterns, predictable from rsfMRI, may relate to behavioral variations.