The effect of spatial smoothing on fMRI decoding of columnar-level organization with linear support vector machine

Area of Science:

• Neuroimaging
• Machine Learning
• Data Analysis

Background:

• Spatial smoothing is often applied in neuroimaging to reduce noise.
• Its effect on multivariate pattern analysis, particularly for decoding, is debated.
• Invertible transformations are assumed to preserve information.

Purpose of the Study:

• To investigate the impact of spatial smoothing on multivariate classification accuracy.
• To determine how spatial smoothing affects the decoding of columnar-level brain organization.
• To analyze the influence of smoothing on generalization scores in support vector machine (SVM) analysis.

Main Methods:

• Theoretical analysis of spatial smoothing's effect on SVM generalization scores.
• Simulations examining sensitivity to input data scaling.
• Application of spatial smoothing in a functional magnetic resonance imaging (fMRI) experiment decoding ocular dominance responses.

Main Results:

• Theoretical analysis showed spatial smoothing can reduce SVM sensitivity to high-frequency patterns.
• In an fMRI study, increased spatial smoothing decreased decoding accuracies.
• Smoothing effects varied across individual subjects, unlike group-level trends.

Conclusions:

• Spatial smoothing can have a substantial impact on decoding performance.
• Informative patterns for columnar decoding are often in higher spatial frequencies on average.
• Individual subjects may exhibit different spatial frequency distributions for informative patterns.