Modern neuroimaging studies frequently combine data collected from multiple scanners and experimental conditions. Such data often contain substantial technical variability associated with image intensity scale (image intensity scales are not the same in different images) and scanner effects (images obtained from different scanners contain substantial technical biases). Here we evaluate and compare results of data analysis methods without any data transformation (RAW), with intensity normalization using RAVEL, with regional harmonization methods using ComBat, and a combination of RAVEL and ComBat. Methods are evaluated on a unique sample of 16 study participants who were scanned on both 1.5T and 3T scanners a few months apart. Neuroradiological evaluation was conducted for 7 different regions of interest (ROI’s) pertinent to Alzheimer’s disease (AD). Cortical measures and results indicate that: (1) RAVEL substantially improved the reproducibility of image intensities; (2) ComBat is preferred over RAVEL and the RAVEL-ComBat combination in terms of regional level harmonization due to more consistent harmonization across subjects and image-derived measures; (3) RAVEL and ComBat substantially reduced bias compared to analysis of RAW images, but RAVEL also resulted in larger variance; and (4) the larger root mean square deviation (RMSD) of RAVEL compared to ComBat is due mainly to its larger variance.

Background Large‐scale multi‐site neuroimaging studies provide higher power for statistical analyses. However, these aggregated datasets are susceptible to unwanted variability resulting from scanner and acquisition protocol differences. To address this problem, a group of intensity normalization and harmonization methods have been developed recently. This study used a paired 1.5T‐3T MRI dataset to evaluate an intensity normalization procedure, RAVEL (Fortin et al., 2016), a data harmonization method, ComBat (Johnson et al., 2007), and a pipeline combining the two. Method Sixteen healthy elderly participants (median 77.5 years (range=70‐79), 25% (n=4) males) underwent 1.5T and 3T T1‐weighted MRI, three months apart. These images (RAW) were used as input for RAVEL (an intensity normalization technique) and/or ComBat (a harmonization procedure) to characterize the cross‐scanner harmonization of FreeSurfer‐extracted summary measures (cortical thickness and/or volume) relevant to Alzheimer’s disease (Zhang et al., 2001). In setting‐1 (RAVEL), measures were extracted from RAVEL‐normalized images. In setting‐2 (ComBat), measures were extracted from RAW images and then harmonized using ComBat. In setting‐3 (RAVEL‐ComBat), ComBat was applied to the measures resulted from setting‐1. Paired t‐tests were used to evaluate cross‐scanner differences of measures in each of these sets. The measures with statistically significant differences were considered with potential scanner differences and decreasing such differences to lose significance, was considered to achieve harmonization. Result Table 1 shows the means of the differences and their 95% CI for statistical significance, indicating 14 measures with potential scanner differences in RAW (blue/yellow under RAW). Out of these 14 measures, RAVEL‐Combat harmonized 13 measures (green under RAVEL‐ComBat) compared to 8 and 9 harmonized measures by RAVEL and ComBat, respectively. For example, for the cortical thickness of the Inferior Temporal Left, the mean of differences for RAW was 0.25mm, 95%CI:(0.16mm, 0.34mm), indicating a statistically significant difference (blue). This difference has been reduced to 0.08mm, 95%CI (‐0.05mm, 0.21mm) using RAVEL‐ComBat (green) and no longer statistically significant. We observed that, both, RAVEL‐ComBat and RAVEL slightly increased the variability of the differences. Conclusion A combination of intensity normalization (RAVEL) and data harmonization (ComBat) performs best in terms of harmonizing data across 1.5T and 3T MRI within the same cohort.