A within-subject quantitative comparison of images from Siemens VE11C and XA30 scanners

2311 

Abstract Submission 

Ross Mair^1,2, Lindsay Hanford^3,2

¹Cemter for Brain Science, Harvard University, Cambridge, MA, ²Martinos Center for Biomedical Imaging, Massachusetts General Hospital, Boston, MA, ³Dept of Pschology, Harvard University, Cambridge, MA

Ross Mair  
Cemter for Brain Science, Harvard University|Martinos Center for Biomedical Imaging, Massachusetts General Hospital
Cambridge, MA|Boston, MA

Lindsay Hanford  
Dept of Pschology, Harvard University|Martinos Center for Biomedical Imaging, Massachusetts General Hospital
Cambridge, MA|Boston, MA

In large human imaging studies with thousands of subjects, longitudinal studies and multi-site trials, both MR scanner stability and consistency of protocol implementation is vital [1]. The software transition on Siemens Magnetom scanners from VE11C, that has been in use since 2016, to the new XA30 platform represents one of the largest changes to the MRI scanner software interface and image handling experienced by the neuroimaging community in over 15 years. Large multi-site studies such as ABCD [2], AABC – the continuation of HCP-Aging [3], and SSBC, are currently grappling with this change. Of particular concern is the impact of changes in software level, and inherent image reconstruction processes on automated MRI-derived measurements of in-vivo human brain volumes from anatomical scans. A few studies have previously probed the repeatability of these measurements, including effects of changes such as scanner vendor, software version, field strength and gradient strength [4,5,6]. In this study, we trialed two short neuroimaging protocols on a small number of subjects scanned on scanners using VE11C and XA30 software with minimal time interval between the scan sessions.

All images were acquired using 3T Siemens MAGNETOM Prismafit MRI scanners (Siemens Healthineers; Erlangen, Germany). In trial one, one subject (F, 30yo) was scanned at McLean Hospital using a scanner running XA30, and the study was repeated at Harvard University the following day, where the scanner used VE11C. 0.8mm isotropic T1w and T2w images were acquired, and a resting-state BOLD scan was acquired at 2.0 mm isotropic resolution, using the multiband-EPI sequences from University of Minnesota. The protocol was replicated exactly on the two scanners. In trial two, two subjects (F, avg 30.0 yo) were scanned at Massachusetts General Hospital using a scanner running XA30, and the sessions were repeated at Harvard University the 4-6 days later, where the scanner used VE11C. These sessions included 1.0mm isotropic T1w and T2w scans with prospective motion correction [7], a resting-state BOLD scan with 2.4 mm isotropic resolution, and a multi-shell diffusion scan at 1.8 mm isotropic resolution and 176 directions. All structural images were processed using MRIQC [8] for estimates of SNR and Image Smoothness, and through FreeSurfer for estimates of subcortical volume and regional cortical thickness. BOLD and diffusion scans were assessed across the software platforms via quantitative metrics from MRIQC or FSL's EddyQuad [9].

Figures 1 and 2 show results from the morphometric analysis of the T1w images from one of the two subjects acquired in the second trial. Figure 1 shows the pial and white-matter surface tracings generated from the robust-registered T1w images acquired using VE11C and XA30 software. The tracings virtually overlay one another. In Figure 2, the correlation of all cortical thickness regions obtained from Freesurfer, and volume measures of sub-cortical structures are shown, for a test-re-test repeatability assessment on the VE11C scanner, and the between-software agreement. The results are virtually identical. Similar results were obtained from the second subject, and the single subject in the initial trial using the 0.8 mm non-motion-corrected T1w protocol. Quantitative analysis of BOLD scans are confounded by effects of subject motion that can vary from session to session even within subjects. While variations were observed in quantitative parameters such as image SNR, image smoothness (FWHM), tSNR and DVARS, the observed between-subject variability was greater than the variability seen across sessions within an individual.

While a very limited study with a small number of subjects, we show that there may be less reason for concern over impacts from the Siemens XA30 software change on quantitative image analysis and metrics than has often been seen in the past as a result of scanner software or hardware changes

Methods Development

Segmentation and Parcellation

Anatomical MRI ¹

BOLD fMRI ²

Data analysis

FUNCTIONAL MRI

MRI

STRUCTURAL MRI