Biomarkers

Area of Science:

• Neuroimaging
• Medical Imaging Analysis
• Biomarker Development

Background:

• The DPUK PET-MR Harmonisation Study addresses variability in brain scans from multiple PET-MR scanners.
• Optimizing clinical trial design and evaluating PET-MR methods are key objectives.
• This research focuses on imaging-derived phenotypes (IDPs) from T1-weighted MRI scans in healthy elderly participants.

Purpose of the Study:

• To quantify within-site and across-site variability in brain scans acquired from different PET-MR scanners.
• To evaluate the effectiveness of various harmonisation methods for T1-weighted MRI scans.
• To improve the consistency of imaging biomarkers for early dementia diagnosis.

Main Methods:

• T1-weighted (T1w) scans were processed using the UK Biobank (UKB) pipeline to extract volumetric IDPs.
• Three harmonisation approaches were assessed: Image Quality Metric (IQM)-based regression, IDP-based ComBat adjustment, and image-based histogram matching or SynthSR.
• Within-subject variability was quantified using the Coefficient of Variation (CoV); cross-subject variability was also assessed.

Main Results:

• Visual inspection revealed scan quality significantly impacts within-subject variability.
• Longitudinal ComBat effectively reduced CoV (e.g., brain volume: 3.7±1.2% to 1.4±0.8%) while preserving cross-subject variability.
• IQM regression reduced variability but also decreased cross-subject variability; image-based methods showed inconsistent or incompatible results.

Conclusions:

• ComBat demonstrated effectiveness in reducing IDP variability but requires specific batch/site variable definitions.
• IQM regression offers flexibility but needs refinement to maintain cross-subject variability.
• Further development is needed for image-based harmonisation methods; quantifying diverse variability aspects is vital for multi-site MRI studies.