Repeatability analysis of ICA-based harmonization for multi-site MRI data using dual projection models.

Area of Science:

• Neuroimaging
• Biomedical Data Analysis
• Statistical Modeling

Background:

• Multi-site magnetic resonance imaging (MRI) data integration boosts research power but introduces site effects that can mask biological signals.
• Independent component analysis (ICA)-based harmonization methods, like dual projection ICA (ICA-DP), aim to reduce site effects while preserving signal.
• The repeatability of ICA decompositions is crucial for reliable site effect removal but remains a significant challenge.

Purpose of the Study:

• To propose a novel evaluation framework for assessing ICA repeatability in multi-site neuroimaging.
• To investigate the role of component energy in ICA stability and harmonization.
• To refine the ICA-DP harmonization scheme for improved signal preservation.

Main Methods:

• Developed a new framework evaluating ICA repeatability by jointly assessing spatial components, mixing coefficients, and component energy.
• Utilized simulated and real multi-site MRI datasets for validation.
• Revised the ICA-DP harmonization scheme to comprehensively remove site-associated components.

Main Results:

• Incorporating component energy into repeatability metrics provides a more robust assessment of ICA stability.
• The proposed framework offers a theoretically grounded evaluation of ICA repeatability.
• The revised ICA-DP harmonization scheme demonstrated improved preservation of biologically relevant signals.

Conclusions:

• Repeatability analysis is essential for reliable ICA-based harmonization in multi-site neuroimaging.
• The proposed framework enhances the assessment of ICA stability and harmonization effectiveness.
• This work provides a reliable tool for mitigating site effects in multi-site MRI studies.