Effects of protocol and scanner changes on segmentation volume estimates in a dedicated crossover data set

Background The Mayo Clinic Study of Aging and Alzheimer’s Disease Research Center used GE scanners exclusively for MRI from 2005‐2017, employing consistent imaging parameters throughout. In 2017 the MRI protocol was updated and migrated to Siemens Prisma scanners; 113 participants were scanned with each configuration. The GE protocol included a 2D FLAIR. The Siemens protocol includes 3D FLAIR and T2‐weighted images. Both vendor protocols included a 3D T1. Experience with multi‐center trials suggests such changes may introduce systematic variations. Methods GE scanning was carried out on 3T scanners with 12 channel phased array head coils. Siemens Prisma scanning was carried out with 64 channel head/neck coils. Imaging sequences and parameters are given in Table 1. We used the MCALT template and priors to perform standard six‐class segmentation on T1‐weighted (T1w) images. SPM12 unified segmentation supports the simultaneous use of multiple input contrasts; segmentation of multi‐contrast inputs was done using eight tissue classes including deep grey and white matter hyperintensity (WMH) priors. For a given MRI study (either GE or Siemens) all images were registered and resampled to the study’s T1w image. Two‐channel (T1+FLAIR) segmentation was performed on GE scans. Two and three‐channel segmentations were performed on Siemens scans. Comparisons included total intracranial volume, total tissue volume and total grey matter volumes, as well as hippocampal volumes and grey matter volume in a temporal lobe meta‐ROI. Results Four participants were excluded for poor image quality: one poor GE T1w quality, one poor Siemens T1, T2 and FLAIR quality, and two poor Siemens T2 quality. Comparisons were made among 109 participants aged 34–97 (Table 2). By visual assessment, adding T2‐weighted images always improved intracranial volume correctness. Using the combination of single‐channel GE and three‐channel Siemens generally improved cross‐vendor agreement on grey matter volumes with the exceptions of hippocampal and intracranial volume estimates. Conclusion Requiring multiple input channels increases the likelihood that the result of a study being excluded. Combining T1w, FLAIR and T2w images allows the delineation of WMH and deep grey matter probability outputs but may introduce differences in GM volumes.