Estimating the effect of a scanner upgrade on measures of grey matter structure for longitudinal designs

Abstract

Longitudinal imaging studies are crucial for advancing the understanding of brain development over the lifespan. Thus, more and more studies acquire imaging data at multiple time points or with long follow-up intervals. In these studies changes to magnetic resonance imaging (MRI) scanners often become inevitable which may decrease the reliability of the MRI assessments and introduce biases. We therefore investigated the difference between MRI scanners with subsequent versions (3 Tesla Siemens Verio vs. Skyra) on the cortical and subcortical measures of grey matter in 116 healthy, young adults using the well-established longitudinal FreeSurfer stream for T1-weighted brain images. We found excellent between-scanner reliability for cortical and subcortical measures of grey matter structure (intra-class correlation coefficient > 0.8). Yet, paired t-tests revealed statistically significant differences in at least 67% of the regions, with percent differences around 2 to 4%, depending on the outcome measure. Offline correction for gradient distortions only slightly reduced these biases. Further, T1-imaging based quality measures reflecting gray-white matter contrast systematically differed between scanners. We conclude that scanner upgrades during a longitudinal study introduce bias in measures of cortical and subcortical grey matter structure. Therefore, before upgrading a MRI scanner during an ongoing study, researchers should prepare to implement an appropriate correction method for these effects.

Same Brain, Different Look?—The Impact of Scanner, Sequence and Preprocessing on Diffusion Imaging Outcome Parameters

“…TBSS on the FA skeleton presented a pattern with higher FA values for Verio data in more superficial white matter and higher FA values for Skyra data in rather deep WM areas. We observed a bias similar to this pattern when comparing anatomical MR images from Verio and Skyra [ 65 ], namely Skyra exhibiting higher cortical thickness and larger GM volumes in medial frontal and central regions and Verio showing higher cortical thickness in lateral and occipital regions. This pattern could be caused by scaling differences between scanners that were observed on the anatomical images and could affect diffusion image processing due to registration on the individual anatomical images.…”

“…Additionally, in a parallel comparison of the anatomical images from Verio and Skyra, gradient non-linearity correction—conducted with the gradunwarp implementation [ (accessed on 5 October 2021)] in Python 2.7. )—did not substantially reduce the detected differences [ 65 ]. This is why we estimate that gradient non-linearities might only explain a small share of the differences in the WM skeletons between the scanners.…”

Adolescent to young adult longitudinal development of subcortical volumes in two European sites with four waves

“…These systematic effects may rather stem from differences in the imaging sequence used on the two sites (higher resolution in Dresden) which may have locally affected the FreeSurfer segmentation algorithm. Further, a scanner update was completed during wave three in Paris (Siemens Trio to Prisma) which may have introduced additional bias in results as Medawar et al (2021) and Plitman et al (2021) reported very good ICC but also percent volume differences in several cortical and subcortical regions when comparing Siemens Scanners (Trio versus Prisma/Verio versus Skyra). Since many recent longitudinal studies on brain development included several samples with differences in imaging sequence and scanner type (Dennison et al, 2013; Mills et al, 2021; Tamnes et al, 2017; Wierenga, Sexton, et al, 2018), results should be interpreted with caution.…”

Harmonized diffusion MRI data and white matter measures from the Adolescent Brain Cognitive Development Study

“…Several studies have revealed that even slight modification to the scanner software can result in bias in the acquired dMRI data, similar to a "site-effect," even when the acquisition protocol remains unchanged (Medawar et al 2021;Bayer et al 2022). However, it is often difficult to predict the bias in the data due to such software updates.…”