INTRODUCTION / PURPOSE
Multicenter research studies face challenges in standardizing imaging protocols, necessitating data harmonization. On a more restricted scale, standardization must be pursued in radiology departments running multiple MRI scanners, to ensure repeatability and consistency of the produced images. This study aimed to quantify and reduce intra-site variability in Diffusion-Weighted (DWI) and structural  MR images.

MATERIALS AND METHODS
Fifty unpaired datasets from two 1.5T MR systems: one standard and one on a mobile unit, were analyzed. Scanners were matched for head coil, pulse sequence parameters, and software version. Region-specific Apparent Diffusion Coefficient (ADC) from DWI and volumetric measurements from T1-weighted images were extracted and statistically compared before and after retrospective harmonization using NeuroComBat.

RESULTS
Morphometry results were similar between scanners, with no appreciable differences. Concerning DWI, retrospective harmonization was effective in controlling significant variability in between-scanners measurements of ADC. Surprisingly, the mobile unit showed reduced sensitivity in detecting age-related changes in ADC, an effect poorly controlled by harmonization.

CONCLUSION
While morphometric data appears to be more robust, intra-site variability affects quantitative metrics such as ADC. Retrospective harmonization effectively mitigated these discrepancies, highlighting its potential to reduce scanner-related variability and ultimately improve the diagnostic performance. We strongly advocate for an extensive and accurate prospective harmonization of MR protocols, especially when quantitative imaging is of interest.

LIMITATIONS
We acknowledge the limited sample size and the exclusion from the analysis of potential confounding factors other than age and gender.