Deconvolving BOLD activation in event-related designs for multivoxel pattern classification analyses

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Machine Learning in Neuroscience

Background:

• Multivoxel pattern analysis (MVPA) is increasingly used in fMRI to predict cognitive states during tasks.
• Calculating activation patterns for rapid event-related designs is challenging due to overlapping BOLD signals.
• Effective MVPA in rapid designs is desirable for increased stimulus presentation and subject focus.

Purpose of the Study:

• To compare eight models for estimating trial-by-trial activation patterns in rapid event-related fMRI designs.
• To identify the most effective method for improving MVPA accuracy in these designs.
• To enable simultaneous optimization of univariate and MVPA approaches in fMRI.

Main Methods:

• Comparison of 8 distinct models for estimating trial-by-trial activation patterns.
• Utilized rapid event-related designs with varying interstimulus intervals and signal-to-noise ratios.
• Employed a general linear model including regressors for individual trials and all other trials.

Main Results:

• The proposed general linear model significantly improved the estimation of trial-by-trial activation patterns compared to standard methods.
• Enhanced accuracy in predicting cognitive states was observed in rapid event-related designs with higher signal-to-noise ratios.
• The method yielded trial-by-trial estimates more representative of true activation magnitudes.

Conclusions:

• A novel general linear model approach enhances MVPA in rapid event-related fMRI designs.
• This method improves the prediction of cognitive states by providing more accurate activation estimates.
• The findings facilitate the simultaneous optimization of univariate and MVPA analyses in fMRI research.