A linked independent component analysis framework for characterizing site-effect patterns in multi-site structural...

Area of Science:

• Neuroimaging
• Biomedical Engineering
• Data Science

Background:

• Multi-site magnetic resonance imaging (MRI) studies offer increased statistical power but face challenges from scanner and protocol variability.
• Existing harmonization methods primarily reduce variance but often overlook the spatial expression and reproducibility of site effects.
• Understanding the specific acquisition parameters contributing to site effects is crucial for improving data quality.

Purpose of the Study:

• To develop and validate a framework for identifying and interpreting site-effect patterns in structural and functional MRI data.
• To assess the spatial reproducibility and technical attribution of identified site effects across different imaging modalities.
• To complement existing harmonization techniques by providing a component-level diagnostic approach for multi-site MRI analysis.

Main Methods:

• Developed a modality-wise Linked Independent Component Analysis (LICA) framework to decompose voxel-wise maps into spatial components and subject-level loadings.
• Analyzed Grey Matter (GM) volume, amplitude of low-frequency fluctuation (ALFF), and regional homogeneity (ReHo) maps separately.
• Classified components based on site labels, biological covariates, spatial reproducibility, and technical attribution using recorded acquisition parameters on ABIDE II data.

Main Results:

• LICA identified distinct site-related components across GM volume, ALFF, and ReHo, demonstrating modality-specific spatial patterns beyond global shifts.
• Grey matter volume exhibited a stable whole-brain site-effect pattern, while functional measures showed more heterogeneous patterns.
• Site labels explained the most variance; acquisition parameters like TR, TE, FA, and voxel size showed modality-dependent contributions to site effects.

Conclusions:

• The LICA framework offers a component-level diagnostic tool for multi-site MRI data analysis.
• Mapping, stabilizing, and interpreting site-effect patterns improves transparency and reproducibility in neuroimaging research.
• This approach complements conventional harmonization methods, facilitating more reliable multi-site structural and functional MRI studies.