Riemannian frameworks for the harmonization of resting-state functional MRI scans.

Area of Science:

• Neuroimaging
• Statistical Analysis
• Data Harmonization

Background:

• Magnetic Resonance Imaging (MRI) provides detailed brain images, but scanner and protocol variations introduce noise.
• Harmonization methods are crucial for reducing variability in neuroimaging datasets, especially with increasing public data availability.
• Resting-state functional MRI (rs-fMRI) presents unique harmonization challenges due to its complex data structure.

Purpose of the Study:

• To develop and validate a novel framework for harmonizing rs-fMRI data.
• To reduce the impact of scanner and protocol differences on functional connectomes and rs-fMRI time series.
• To preserve essential clinical information related to aging and brain disorders during harmonization.

Main Methods:

• Proposed a harmonization approach using Riemannian geometric frameworks to maintain functional connectome properties.
• Integrated state-of-the-art harmonization techniques within the Riemannian framework.
• Generated and analyzed synthetic data to compare eighty variants of the proposed method.
• Applied the best-performing framework to real-world datasets (ABIDE, HCP, Framingham, GenStruct).

Main Results:

• The proposed framework successfully harmonized low-dimensional connectomes and voxelwise functional time series.
• Demonstrated the effectiveness of preserving connectome properties during the harmonization process.
• Validated the approach on diverse datasets, including low and high-dimensional rs-fMRI data.

Conclusions:

• The developed Riemannian-based framework offers effective harmonization for rs-fMRI data.
• Preserving the mathematical properties of functional connectomes is essential for successful harmonization.
• This method enhances data consistency and reliability for neuroimaging research.