A resource for development and comparison of multimodal brain 3 T MRI harmonisation approaches

Area of Science:

• Neuroimaging
• Quantitative Magnetic Resonance Imaging (qMRI)
• Data Harmonization

Background:

• Magnetic resonance imaging (MRI) holds significant potential for brain mapping but faces limitations in consistency, reproducibility, and accuracy.
• Non-biological factors like hardware, software, and calibration differences introduce variability, hindering quantitative MRI (qMRI) and interfering with biological signal detection.
• Lack of harmonization across neuroimaging datasets is a major challenge for reliable qMRI.

Purpose of the Study:

• To create a comprehensive resource for mapping the extent of inconsistency in neuroimaging data.
• To objectively evaluate different neuroimaging harmonization approaches.
• To identify optimal processing strategies and features for minimizing between-scanner variability in qMRI.

Main Methods:

• Utilized a traveling-heads paradigm with 10 subjects scanned across 5 sites, 6 scanners (3 major vendors), and 5 MRI modalities.
• Acquired multiple within-scanner repeats to establish scan-rescan variability baselines.
• Extracted hundreds of imaging-derived phenotypes to compare between-scanner, within-scanner, and biological variability.

Main Results:

• Characterized feature reliability across scanners and identified optimal processing pipelines for implicit harmonization.
• Evaluated explicit harmonization tools, assessing their efficiency in reducing between-scanner variability.
• The ON-Harmony dataset provides a benchmark for multimodal scan-rescan variability and between-scanner differences.

Conclusions:

• Developed the ON-Harmony dataset, a valuable resource for understanding and addressing qMRI variability.
• Provided good practice suggestions for processing steps and feature selection to enhance consistency.
• Established references for future qMRI harmonization studies and highlighted the need for robust harmonization strategies.