Harmonization of diffusion MRI data sets with adaptive dictionary learning.

Area of Science:

• Medical Imaging
• Biophysics
• Computational Neuroscience

Background:

• Diffusion magnetic resonance imaging (dMRI) provides insights into tissue microstructure.
• Scanner hardware variations introduce unwanted variability, confounding biological interpretations.
• Harmonization is crucial for multicenter studies to ensure data consistency.

Purpose of the Study:

• To develop a novel harmonization algorithm for dMRI data.
• To mitigate scanner-induced variability while preserving biological differences.
• To improve data comparability across different MRI scanners and protocols.

Main Methods:

• Proposed an adaptive dictionary learning algorithm for dMRI harmonization.
• Learned overcomplete dictionaries iteratively from all datasets simultaneously.
• Applied a scanner-space mapping directly in native subject space without paired data or registration.

Main Results:

• Successfully removed scanner-specific variability from dMRI metrics.
• Preserved the effect size of four key diffusion metrics.
• Demonstrated preservation of biologically relevant alterations (e.g., free water compartment) post-harmonization.
• Reduced global data variability while maintaining biological signals.

Conclusions:

• The adaptive dictionary learning algorithm effectively harmonizes dMRI data across scanners.
• This method facilitates multicenter studies by enabling data pooling and increasing statistical power.
• The approach preserves biologically relevant information, crucial for clinical and research applications.