Principal component regression predicts functional responses across individuals.

Area of Science:

• Neuroimaging analysis
• Statistical inference
• Brain imaging

Background:

• Inter-subject variability poses a significant challenge for group-level neuroimaging studies.
• Standard analysis models struggle to capture complex between-subject patterns, reducing statistical sensitivity.

Purpose of the Study:

• To introduce a novel analysis framework to model and reduce inter-subject variability in neuroimaging.
• To leverage information from multiple imaging contrasts to improve group-level inference.

Main Methods:

• Developed a framework to estimate variance explained by a random effects subspace learned across multiple images.
• Utilized principal component regression as the core estimator within the framework.

Main Results:

• Principal component regression demonstrated superior performance compared to other regression models.
• The proposed method successfully accounted for a significant proportion (10% to 25%) of between-subject variability.
• This validates the utility of accumulating information from multiple contrasts.

Conclusions:

• The novel framework effectively models random subject effects by utilizing redundant information across imaging contrasts.
• This approach enhances the sensitivity of neuroimaging group analyses by addressing inter-subject variability.
• Accumulating contrasts within individuals provides a basis for more robust group-level inference.