Harmonization of multi-site functional MRI data with dual-projection based ICA model

Area of Science:

• Neuroimaging
• Data Harmonization
• Biomedical Data Analysis

Background:

• Multi-site neuroimaging studies enhance statistical power and generalizability.
• Site-specific scanner variations introduce measurement biases, masking true biological signals.
• Harmonizing data across sites is crucial for reliable neuroimaging research.

Purpose of the Study:

• To evaluate the efficacy of a dual-projection (DP) based independent component analysis (ICA) denoising method for harmonizing multi-site functional magnetic resonance imaging (fMRI) data.
• To assess the method's ability to remove site effects from fMRI data, specifically amplitude of low frequency fluctuation (ALFF) and regional homogeneity (ReHo) measures.
• To determine if the DP-based ICA denoising improves the detection of associations between non-imaging variables and fMRI measures.

Main Methods:

• Application of a dual-projection (DP) based independent component analysis (ICA) denoising strategy.
• Analysis of functional magnetic resonance imaging (fMRI) data from the Autism Brain Imaging Data Exchange II (ABIDE II) dataset.
• Validation using frequency-domain analysis, regional homogeneity (ReHo), and amplitude of low frequency fluctuation (ALFF) metrics.

Main Results:

• The DP-based ICA denoising method effectively removed site-related artifacts from multi-site fMRI data.
• Harmonization using the DP method led to increased statistical significance in associations between non-imaging variables (e.g., age, sex) and fMRI measures.
• The method demonstrated successful application to both ALFF and ReHo modalities.

Conclusions:

• The DP-based ICA denoising method is a viable strategy for harmonizing fMRI data in multi-site neuroimaging studies.
• This approach enhances the accuracy and reliability of findings by mitigating scanner-induced biases.
• The method facilitates more robust and reproducible neuroimaging research by enabling effective data pooling.