Shared component analysis.

Area of Science:

• Neuroimaging
• Computational Neuroscience
• Data Science

Background:

• Increasing neuroimaging data complexity necessitates advanced dimensionality reduction techniques.
• Principal Component Analysis (PCA) assumes variance equals relevance, which may not hold for weak but important neuroimaging signals.
• Overfitting and high computational costs are challenges with large-scale neuroimaging datasets.

Purpose of the Study:

• To introduce Shared Component Analysis (SCA) as a novel method for neuroimaging data dimensionality reduction.
• To provide an alternative to PCA that prioritizes components shared across multiple sensors.
• To address the limitations of PCA in identifying relevant but low-variance sources in neuroimaging data.

Main Methods:

• Shared Component Analysis (SCA) algorithm development.
• Iterative application of normalization and PCA.
• Linear data transformation to prioritize widely shared components.
• Evaluation of SCA's effectiveness in neuroimaging data analysis.

Main Results:

• SCA identifies components that are consistently present across multiple sensors.
• The method prioritizes components potentially originating from deep brain sources.
• SCA serves as a plug-in replacement for PCA in dimensionality reduction tasks.

Conclusions:

• SCA is a viable alternative to PCA for neuroimaging dimensionality reduction.
• The method's focus on shared components offers advantages for detecting subtle but relevant neural signals.
• SCA contributes to more effective analysis of complex, high-resolution neuroimaging data.