Group linear non-Gaussian component analysis with applications to neuroimaging.

Area of Science:

• Neuroimaging
• Machine Learning
• Biomarker Discovery

Background:

• Functional magnetic resonance imaging (fMRI) is widely used for neurological disorder research.
• Group Independent Component Analysis (ICA) is a common method but may discard low-variance features.
• Principal Component Analysis (PCA) preprocessing in Group ICA can limit biomarker detection.

Purpose of the Study:

• To introduce a novel group Linear Non-Gaussian Component Analysis (LNGCA) model for fMRI data.
• To enable simultaneous dimension reduction and feature estimation, including low-variance features.
• To develop a method for extracting both shared group components and subject-specific components.

Main Methods:

• A group LNGCA model was developed to extract shared and individual components from fMRI data.
• A parametric resampling test was used to determine the number of components per subject.
• The model was applied to resting-state fMRI data from children with and without autism spectrum disorder.

Main Results:

• Simulations demonstrated higher accuracy of group LNGCA compared to group ICA.
• The method successfully identified resting-state networks in the autism spectrum disorder cohort.
• Group components showed differential temporal engagement patterns between typically developing children and those with autism.

Conclusions:

• Group LNGCA provides a promising new approach for feature detection in neuroimaging.
• This method enhances the identification of subtle differences in brain activity related to neurological disorders.
• The technique offers improved accuracy and retains valuable low-variance information for biomarker discovery.