Structural and functional multi-platform MRI series of a single human volunteer over more than fifteen years

We present MRI data from a single human volunteer consisting in over 599 multi-contrast MR images (T1-weighted, T2-weighted, proton density, fluid-attenuated inversion recovery, T2* gradient-echo, diffusion, susceptibility-weighted, arterial-spin labelled, and resting state BOLD functional connectivity imaging) acquired in over 73 sessions on 36 different scanners (13 models, three manufacturers) over the course of 15+ years (cf. Data records). Data included planned data collection acquired within the Consortium pour l’identification précoce de la maladie Alzheimer - Québec (CIMA-Q) and Canadian Consortium on Neurodegeneration in Aging (CCNA) studies, as well as opportunistic data collection from various protocols. These multiple within- and between-centre scans over a substantial time course of a single, cognitively healthy volunteer can be useful to answer a number of methodological questions of interest to the community.


Background & Summary
Reproducibility is a cornerstone scientific principle, especially so when one is faced with measuring instruments that do not produce absolute quantifiable data such as magnetic resonance imaging (MRI). The resulting signal varies systemically based on either the type and parameters of the acquisition being chosen, as well as randomly due to the instrument itself. These effects produce changes in signal (i.e. images) that may reach statistical significance when compared to the effect under study. Calibration is therefore needed.
Geometric objects ("phantoms") have been proposed to track and correct some of these effects, and are presently in routine use. However, there is no geometric phantom however is able to approximate with sufficient fidelity the intricacies of living organs, especially in the case of the human brain, nor mimic its physiological characteristics. To this end, a number of studies have proposed and are using human volunteer(s) as calibration devices, postulating that the time course for changes in any one of the volunteers' brain will be of smaller magnitude than the changes being measured; in effect, that these brains can be considered nearly identical to themselves across time (the length of the study) and space (scanners and sites).
There have been many instances of human volunteer brain data made accessible for public use. One of the earliest known is referred to as Colin27 1,2 and consisted in 27 different T1-, T2-and proton-density weighted (T1w, T2w, PD) MRI acquisitions of a single individual, acquired on a single scanner within weeks (http://www.bic. mni.mcgill.ca/ServicesAtlases/Colin27). Table 1 details other works that have been done since, by Brown et al. 3 , Friedman et al. 4 , Han et al. 5 , Suckling et al. 6 , and in these pages, efforts coordinated via the Consortium for Data reliability and reproducibility (https://doi.org/10.15387/fcp_indi.retro.simon) by Maclaren et al. 7 , and Wakeman et al. 8 . While extremely useful, one notices a need for a dataset of individual(s) with longitudinal acquisitions of a multimodal protocol using a variety of scanners models and vendors. Such a dataset would serve the needs of multiple imaging researchers to test the reproducibility and robustness of their methods.
Our goal was to fill this gap by contributing data from a single human volunteer originating from two settings 9 . In the first, a planned data collection is reported as the single participant acted as a human volunteer for a number of multi-centric studies that began coordinated data acquisition using a trans-national harmonized protocol. The second is an opportunistic data collection with various protocols, as this same volunteer participated in a number of unrelated academic studies, protocol tests, and manufacturer visits (2001-present). Hence this second collection was acquired in a disharmonized fashion. In fine, the dataset consists of over 599 multi-contrast MR images (T1w, T2w, PD, fluid-attenuated inversion recovery (FLAIR), gradient-echo (T2*), diffusion-weighted (DWI), susceptibility-weighted (SWI), arterial-spin labeling (ASL), time-of-flight (ToF) and resting state BOLD functional connectivity imaging (rsfMRI). This dataset was acquired through 73 sessions on 36 different scanners (13 models, three manufacturers) over the course of the last 17 years (cf. Data records). Given these multiple withinand between-centre scans over a substantial time course, we propose that this data can be useful in answering a number of methodological questions of interest to the community (cf. Usage Notes).

Participant and ethical agreement. A Single Individual volunteering for Multiple Observations across
Networks (SIMON) took part in these studies. SIMON is an ambidextrous male, aged between 28 to 46 years old across the timeline of the dataset, and remains free from known cognitive impairment, neurological disease or psychiatric disorders. Ethical agreements for the purposive acquisitions were obtained from the coordinating ethics committee of either the Institut universitaire de gériatrie de Montréal (Consortium pour l'identification précoce de la maladie d'Alzheimer -Québec study (CIMA-Q)); the Lady Davis Institute (Canadian consortium for neurodegeneration and aging study (CCNA)); or of the Ontario Brain Institute's Ontario Neurodegenerative Disease Research Initiative (ONDRI) 10 . Ethical agreements for the convenience acquisitions were obtained by each individual investigator at their respective institutions. The participant gave his written informed consent before enrollment in each study and has consented to release his data without restriction.
Part of the data used in this article were obtained from the Consortium pour l'identification précoce de la maladie Alzheimer -Québec (CIMA-Q), founded in 2013 with a $2,500,000 grant from the Fonds d'Innovation Pfizer -Fond de Recherche Québec -Santé sur la maladie d'Alzheimer et les maladies apparentées. The main objective is to build a cohort of participants characterized in terms of cognition, neuroimaging and clinical outcomes in order to acquire biological samples allowing (1) to establish early diagnoses of Alzheimer's disease, (2) to provide a well characterized cohort and (3) to identify new therapeutic targets. The principal investigator and director of CIMA-Q is Dr Sylvie Belleville of the Centre de recherche de l'Institut universitaire de gériatrie de Montréal, CIUSSS Centre-sud-de-l'Île-de-Montréal. CIMA-Q represent a common effort of several researchers from Québec affiliated with Université Laval, McGill University, Université de Montréal, and Université de Sherbrooke. CIMA-Q recruited 350 cognitively healthy participants, with subjective cognitive impairment, mild cognitive impairment, or Alzheimer's disease, between 2013-2016.
Purposive acquisitions -MRIs from the canadian dementia imaging protocol. In Canada, a collection of studies centred on dementias started acquiring data in the 2013-2017 timeframe which required a common, harmonized process for MRI data acquisition. This collection defined what is referred to as the Canadian Dementia Imaging Protocol (CDIP; www.cdip-pcid.ca) 11 . The CDIP ( Fig. 1) includes the following sequences: (a) an isotropic high-resolution 3D T1w MRI; (b) an interleaved PD/T2w MRI; (c) T2* and FLAIR MRIs; (d) a 30+ directions DWI MRI; and (e) a functional connectivity, resting-state BOLD MRI. Parameters for each one of these sequences were harmonized across three MRI vendors, namely General Electric Healthcare, Phillips Medical Systems, and Siemens Healthcare. Sites were evaluated to be CDIP-compliant when they followed a three-step process, namely qualification of the protocol, on-going quality control using geometric and human phantom scans, and on-going quality assurance during the study.
Within this context, SIMON was therefore scanned at 20 different sites in the 2014-2018 time period using this harmonized protocol, namely at the Brain Imaging Center of the Institut universitaire en santé mentale Douglas (Montreal, Canada), the Brain Imaging Center of the Montreal Neurological Institute (Montreal, Canada), the Center for Addiction and Mental Health (Toronto, Canada), the Centre d'Imagerie Moléculaire de Sherbrooke of the Centre Hospitalier Universitaire de Sherbrooke (Sherbrooke, Canada), the Centre Hospitalier Universitaire de

Data Records
Characteristics of the MRIs are described in Table 2 for MRIs acquired with the CDIP, and Table 3   www.nature.com/scientificdata www.nature.com/scientificdata/ technical Validation Qualitative control. All images have been reviewed for quality and, when applicable, conformance to the CDIP acquisition parameter values. This review verified the following: coverage, presence of artefacts, and adherence to protocol parameters for purposive acquisitions. Three T1w, one T1-Cube, and one T2-Cube images from the GE manufacture are missing small parts of the right portion of the brain. These images were not included in the analyses.
Quantitative control. For the quantitatively assess image quality, we calculated the contrast to noise ratio (CNR) for the single individual volunteer using voxel intensities from raw T1w images with the mean of grey matter (GM) and cerebral white matter (WM): This CNR definition was chosen since as it is simple and frequently used 11,12 . Subcortical and cortical GM and WM volumes were obtained from the aparc and aseg labels derived from automated segmentation using FreeSurfer (version 5.3 with default parameters; freesurfer.net). Please note that FreeSurfer segmentations are not distributed with the dataset. The technical details of FreeSurfer procedures are described in prior publications [13][14][15][16][17] . CNR was computed using the uncorrected T1w image (orig.mgz file). Supplementary In order to compute CNR for FLAIR images, the same equation as T1w was used. However, since there is little contrast between WM and GM classes in FLAIR, the first tissue class is called the brain class (which is GM + WM as a whole), and the second tissue class is the CSF. In order to compute the mean and standard deviations for the brain and CSF tissues, segmentations of these tissues were required. Tissue segmentation was generated using a standardization and segmentation framework for FLAIR MRI [18][19][20] . To ensure that the tissue classes contained pure tissues only, 70% of the middle slices were retained for CNR calculation, ensuring that if the brain  www.nature.com/scientificdata www.nature.com/scientificdata/ extraction algorithm missed any skull at the top or bottom, it would not bias the approach. Results are shown in Supplementary Table 3 and Fig. 3 and show that FLAIR CNR significantly differs between vendors (H: 50.83, p < 0.0001, GE: 346.17 ± 0.00, Philips: 1270.08 ± 326.33, Siemens: 340.95 ± 102.71), with Siemens having lower CNR than Philips (U: 0, p < 0.0001) and GE (U: 32, p < 0.0001) manufacturers.
Although, the participant is a healthy individual, white matter hyperintensity (WMH) on FLAIR was assessed using Schmidt et al. 's automated VBM toolbox 21 (GE: 0.55 ml ± 0.00, Philips: 0.33 ml ± 0.21, Siemens: 0.59 ml ± 0.19). Figure 4 shows that the volume of WMH was larger (U:, 177, p < 0.0001) on Siemens than on Philips scanners (see all results in Supplementary Table 2).
Whole-brain tractography was done using the TractoFlow pipeline 22,23 . Figure 5 displays the mean fractional anisotropy (FA) of a tract section located in the mid-body of the corpus callosum. Based on 28 DWI scans from 16 different scanners, this pipeline is dividing brain white matter into 12 bilateral and 9 commissural tracts. The mean FA value obtained from this tract displays high agreement and shows large SD as voxels located in the central core of the corpus callosum have FA values ~0.8 while voxels near the end of the fibers and at crossing fibers have lower FA values.
Stability of resting-state networks across time and vendors were assessed using 24 scans (from the planned data collection), acquired at 13 sites, on three scanner vendors 24 . In brief, to assess the stability of resting-state networks, voxel-wise connectivity maps associated from seven MIST atlas network templates 25 were computed for each rsfMRI scan using NIAK's connectome pipeline 26 (http://niak.simexp-lab.org/pipe_connectome.html). For each of the seven rsfMRI networks, a scan by scan similarity (Pearson's correlation) matrix was generated to summarize the consistency of connectivity maps across the 24 scans. Figure 6 shows the average connectivity for the default-mode network (DMN) for the 24 scans, along with the intra-site (for the six sites with 2 or more scans), intra-vendor and inter-vendor consistency of DMN connectivity over time. Moreover, a series of explanatory variables (time between scans, intra-vendor, inter-vendor, intercept) were assembled for a general linear model analysis. A linear mixture of the explanatory variables were adjusted on the inter-scan consistency measures (dependent variable) using ordinary least squares, for each network separately. For each network, the significance of the effect of inter-vendor, intra-vendor, intra-site and time were tested.

Usage Notes
This dataset 9 , composed of a single individual volunteer data, is useful for testing cross-sectional and longitudinal stability of image processing algorithms across scanner models and vendors. To assess variability across sites in multi-centric studies, such data are required and are relatively rare, and their availability is of paramount importance to the field of image processing. Researchers can compare variance within and between techniques,  Table 3. Opportunistic sample characteristics. www.nature.com/scientificdata www.nature.com/scientificdata/   www.nature.com/scientificdata www.nature.com/scientificdata/ homogeneity, influence of parameters on final measurements, or the effect of various corrective techniques. While most scans are relatively recent, one should note that images in the dataset were acquired when the subject was between 28 to 46 years old and thus, age need to be taken into account for measurement reliability analyses.
The CDIP protocol is already in use as the core protocol for the CIMA-Q study (143 participants; seven sites), and the Ontario Brain Institute's Ontario Neurodegenerative Disease Research Initiative (~600 participants, 11 sites). It serves as the basis for the core protocol of the Canadian Alliance for Health Hearts and Minds, whereas the core protocol (~7,000 participants) includes both T1-weighted and FLAIR acquisitions, and the extended version is the complete CDIP (~1,500 participants; eight sites). It has been deployed at more than 19 sites in Canada as part of the CCNA (~1,600 participants). Together, these studies plan on recruiting well over 10,000 participants, across adulthood and a range of cognitive conditions. The protocol has now reached beyond Canadian shores and is being used at the Human Brain Mapping Unit (Cuban Neuroscience Center, Habana, Cuba) and the University of Electronic Science and Technology of China (Chengdu, China).
A website has been established (www.cdip-pcid.ca) to collect all information regarding the protocol. Dissemination is free and only requires acknowledging the current effort. On the site, visitors can find (a) the full protocol, with all harmonized parameters detailed for each vendor; (b) exam cards or PDF output for all tested scanners to date from the three major vendors, ready for upload in a similar machine; (c) a complete operator manual for MR technologists, complete with descriptions of the procedure to scan the geometric phantom, the single individual volunteer, and any study participant, with a list of common acquisition artefacts and pointers to correct them; and (d) the SIMON datasets.

Code availability
No custom code was used to produce the dataset.