Intensity warping for multisite MRI harmonization

Abstract

In multisite neuroimaging studies there is often unwanted technical variation across scanners and sites. These "scanner effects" can hinder detection of biological features of interest, produce inconsistent results, and lead to spurious associations. We propose mica (multisite image harmonization by cumulative distribution function alignment), a tool to harmonize images taken on different scanners by identifying and removing within-subject scanner effects. Our goals in the present study were to (1) establish a method that removes scanner effects by leveraging multiple scans collected on the same subject, and, building on this, (2) develop a technique to quantify scanner effects in large multisite studies so these can be reduced as a preprocessing step. We illustrate scanner effects in a brain MRI study in which the same subject was measured twice on seven scanners, and assess our method's performance in a second study in which ten subjects were scanned on two machines. We found that unharmonized images were highly variable across site and scanner type, and our method effectively removed this variability by aligning intensity distributions. We further studied the ability to predict image harmonization results for a scan taken on an existing subject at a new site using cross-validation.

Keywords: Elsarticle.cls; Image harmonization; Intensity normalization; Multisite imaging; Warping.

Figure 1:

Histograms of voxel intensities for scan-rescan data on a single subject with MS across seven sites in the NAIMS pilot study: Brigham and Women’s Hospital (Brigham), Cedars-Sinai, Johns Hopkins University (JHU), National Institutes of Health (NIH), Oregon Health & Sciences University (OHSU), University of California San Francisco (UCSF), and Yale University (Yale). Left panel shows raw voxel intensities; right panel shows densities after mica harmonization and White Stripe normalization. At each site two scans were collected; a 1 or 2 after site name indicates the first or second scan, respectively.

Figure 2:

Estimated T2 lesion volumes for scan-rescan pairs at each of 7 sites in the NAIMS study. Circles indicate scan 1 and triangles indicate scan 2. Light and dark colors are volumes for White Stripe normalized images and mica normalized images, respectively.

Figure 3:

CDFs of intensities before and after harmonization by tissue type in the trio2prisma study. Rows indicate tissue type, with whole brain, white matter, and gray matter shown in rows 1, 2, and 3, respectively. Columns correspond to different harmonization methods.

Figure 4:

Boxplots of pairwise Hellinger distances across all subjects, shaded by method. Columns show results for full brain (left), white matter (middle), and gray matter (right). Pairwise distances for Prisma and Trio scans are included for reference.

Figure 5:

Axial slice of skull-stripped, T1-weighted images from a single subject in the trio2prisma dataset. At left is an image collected on the Prisma scanner. The top row from left to right show an image collected on the Trio scanner that has been spatially registered to the Prisma image, the Trio image after mica harmonization, and the Trio image after histogram matching, respectively. The bottom row shows image residuals indicating the voxelwise differences between the Prisma image and the Trio, mica harmonized, and histogram matching normalized images, respectively.

Figure 6:

Left Panel: boxplots of normalized root mean square voxel-wise error for trio2prisma image pairs across methods. Shown are raw image pairs (raw: median interquartile range = 0.3998 {0.0545}), histogram matched pairs (hm: 0.1003 {0.0055}), White Stripe normalized pairs (ws: 0.0428 {0.0052}), mica harmonized pairs (mica: 0.0394 {0.0051}), and pairs harmonized using the leave-one-scan-out approach (mica-loso: 0.0410 {0.0049}). Center and Right Panels: boxplots of difference in Trio and Prisma volumes of white matter (WM) and gray matter (GM), where WM/GM segmentation was performed after normalization/harmonization for White Stripe and mica, respectively.