Optimising Analysis Choices for Multivariate Decoding: Creating Pseudotrials Using Trial Averaging and Resampling

Area of Science:

• Neuroimaging
• Cognitive Neuroscience
• Machine Learning in Neuroscience

Background:

• Multivariate pattern analysis (MVPA) is crucial for identifying condition-specific activation patterns in neuroimaging data.
• MVPA decoding can detect subtle information missed by univariate methods.
• Analysis choices, like trial averaging, significantly impact MVPA decoding outcomes.

Purpose of the Study:

• To systematically evaluate the influence of trial averaging and resampling on MVPA decoding accuracy and statistical significance.
• To determine optimal parameters for trial averaging and resampling in neuroimaging data analysis.
• To provide researchers with tools for optimizing these parameters.

Main Methods:

• Simulated neuroimaging data using CoSMoMVPA and SEREEGA toolboxes.
• Systematic assessment of varying degrees of trial averaging (percentage of total trials).
• Evaluation of the impact of resampling on decoding accuracy and statistical outcomes.

Main Results:

• Modest trial averaging (5%-10% of trials per condition) improved decoding accuracy and t-statistics.
• Resampling offered benefits to t-statistics and classification performance but was not universally required.
• Optimal parameters were dependent on the specific classifier and cross-validation strategy.

Conclusions:

• Judicious trial averaging and resampling are key to enhancing MVPA performance in neuroimaging.
• The study provides empirical evidence for optimizing analysis choices in MVPA.
• Researchers can use the provided code to tailor MVPA parameters to their specific data.