For Magnetic Resonance Imaging (MRI), artefacts are most frequently caused by patient head motion due to long acquisition times of typically 30–60 min (1). These artefacts manifest as ghosting, blurring or signal variations, thus reducing overall image quality and resulting in unsuccessful diagnoses (2). Andre et al. (3) determined the percentage of at least partly repeated MRI examinations to approximately 20%, leading to an estimated increased cost of 115,000 US dollars per scanner per year in the US due to motion.

Children tend to move more in the MR scanner than adults since they are intrinsically more reluctant to lie still and are more negatively influenced by the large size of the scanner, the narrow bore and the loud noises during image acquisition (4). So far, motion artefacts for children between 4 and 10 years are mostly reduced by sedation or general anaesthesia (GA), with different GA rates reported for different age ranges: 61–100% for children between 4 and 6 years (5), 80% for children aged 1-6 years (6) and 40% for children aged 7–12 years (6). However, the need for an anaesthetist to administer the drug and monitor the child, increases the patient's waiting times, as well as the costs of the examination. Slipsager et al. (7) calculated the additional cost for using GA in MRI examinations to 319,000 US dollars per scanner per year in Denmark. Furthermore, there are concerns about adverse events like airway obstruction or oxygen desaturation of GA in young children (1, 4, 8, 9).

Currently, different approaches for preventing motion artefacts without general anaesthesia or sedation are under research. These include strategies for preparing the children with story books and mock scanners before they undergo MRI and distracting them visually and acoustically during the examination (4, 5, 10–12). Another approach is to apply motion correction during or after image acquisition, for which a variety of methods have been developed and tested. These include prospective motion correction updating the position of the field of view in real time dependent on motion estimates (13–15), as well as several retrospective techniques using mathematical properties of the Fourier Transform, Compressed Sensing or Machine Learning (16–19).

For both strategies—preparation and motion correction—information about the children's motion patterns in the MR scanner is essential. In the case of motion correction algorithms, additional information is needed for tailoring them toward children specific movements. For the preparation strategy as well as in the clinical setting, information about predominant motion patterns is critical for preventing those types of motion. Churchill et al. (20) analysed head motion of adults during fMRI using retrospective image-based motion correction. Overall, they found low motion estimates with only two cases of estimates above 1mm or 1° for the whole brain. The largest standard deviation of the displacement relative to a reference image was observed for pitch movement (nodding). Afacan et al. (1) investigated children's head motion and its impact on image quality. Their analysis showed a correlation of mean displacement and motion free time to image quality, but no statistically significant correlation of maximum displacement. Additionally, they did not find a significant correlation between the analysed motion metrics and age. Together, these results suggest that head motion is a challenge, not only for young and uncooperative children.

The aim of this work is to analyse motion patterns of children in order to draw possible conclusions about predominant movement habits and thus, enable optimisation of motion correction methods for specific types of motion. For that, we also investigate how head motion translates into motion of different parts of the brain and whether different regions in the brain are affected more or less severely by the children's motion. The analyses will be performed for children imaged with and without GA.

We analysed motion data from anaesthetised and awake children during MRI examination using the same MR scanner and a variable set of sequences from the same base protocol for all patients, which to our knowledge has not been reported so far. Please note that the order and specific parameters of the sequences varied from patient to patient, since the data was acquired in a clinical study.

Comparing motion metrics like mean, median, maximum displacement and motion-free time in Figure 5, confirmed our expectations that children without GA moved significantly more than children with GA for all analysed metrics. Nevertheless, residual movement was still observed for anaesthetised children. As described in the Introduction, Afacan et al. (1) investigated the correlation of motion metrics like mean and maximum displacement and motion-free time with image quality based on motion estimates, which were measured with two sensors placed on the patient's forehead. For enabling comparisons with their work, we calculated our metrics for the point cloud centroid, which corresponds to a point close to the middle of the nose bridge, as visualised in Supplementary Figure 3. Even though Afacan et al. did not find significant differences between radiologic evaluation and maximum displacement, we reported maximum movement since it provides valuable information about the severity of motion spikes, especially in the context of motion correction.

Furthermore, in Figure 5, we showed across all metrics that even anaesthetised children moved: with mean displacements of up to 1.9 mm, maximum displacements of up to 4.4 mm and motion-free times down to 58 %. Afacan et al. (1), who analysed the correlation of motion metrics with radiologic evaluation of image quality, reported motion-free time medians of 38%± 21, 74%± 27, 97%± 17 and 99%± 3 for the grades 1, 2, 3, and 4, respectively (4 being the best image quality). Together with our results of motion-free time values down to 58%, this indicates that GA does not guarantee high image quality due to considerable residual motion in several cases. Our different threshold for calculating motion-free times, does not falsify these conclusions, since lowering our threshold from 2 to 0.2 mm as in (1)] would only decrease our percentages of motion free times. Please note that Afacan et al. also used a different motion tracking device, an electromagnetic tracker (27) developed by Robin Medical Inc. (Baltimore, MD). For other motion tracking devices using external fitted tools, like the Polaris Vicra, inaccuracy in the motion estimates has been shown due to dislocation of the tool relative to the head (28).

In order to obtain more insight into which types of movement children perform most commonly during MR examinations, we decomposed the transformation matrices for each time point into translation and rotation. For both patient groups, we compared the components along / around the three axes. In Figure 6, we showed for both patient groups significantly larger absolute translation along the z-axis compared to x- and y-axes. For the children without GA, we additionally observed significantly larger rotations around the x-axis compared to the y- and z-axes. The z-translation, which corresponds to translation along the axis through the scanner bore, can be interpreted as a drift motion, especially for the children with GA. This becomes apparent when looking at the complete motion data throughout the whole scan session e.g., in Figure 4. For most of the GA patients one can observe a continuous increase in the displacement along the z-axis from the beginning until the end of the examination. Since we observed negative z-translation values for both patient groups, this motion corresponds to a gliding downwards out of the scanner and could be explained both by relaxation of the child's neck muscles, as well as compression of the foam padding which the head lies on. The rotational component around the x-axis observed for children without GA corresponds to a nodding motion, which together with a sliding in negative z-direction has previously also been observed as primary motion for adults in the MR scanner (20, 29). For children specifically, this movement can be explained by parents or a screen with a movie being positioned at the end of the scanner bore. Thus, our analysis provides important conclusions for clinical examinations of children, namely that avoiding nodding motion should play a larger role in training children before an examination.

Moreover, we analysed how the children's motion affects different parts of the brain. For this, we chose 8 cortical and 8 subcortical regions of the brain. We calculated the displacement along each axis for the selected regions. We did not observe a difference when comparing the displacements between regions. The region-wise results were also in accordance with the results from the matrix decomposition presented above. For the GA group very little motion was observed on the x- and y-axes. The group without GA showed more motion on the x- and y-axes, but the z-axis was still dominating. The differences were statistically significant for all regions, except the precentral region of the right hemisphere for the patient group without GA. For this region, the median motion was more distributed across y- and z-axes.

Our study is not without limitations. First, it is limited by the unbalanced distribution between the groups with and without GA. A larger GA group might enable higher statistical significance for comparing the motion metrics between both groups. Second, the examined group of children only consisted of patients with brain tumors. Other patient groups are to be expected to show different motion patterns, for instance some diseases, such as epilepsy or cerebral palsy, come along with an increased tendency for motion. Third, no single motion metric used in the analysis is suitable for summarising the whole motion pattern. For instance a short, but severe nodding increases the maximum displacement, even though the child could have potentially remained motionless for the rest of the scan—with little effect to image quality. However, the combination of several motion metrics (mean, median, and maximum displacement, as well as motion-free time) allows a more complete evaluation of the child's motion, since the different metrics pick up different motion patterns. Furthermore, we only analysed motion patterns and metrics across the whole protocol, which enables us to draw conclusions about which motion children are expected to perform during a whole MR examination. However, we cannot draw conclusions about the motion characteristics and with it, the image quality of individual sequences. Another potential source of error is the fact that FreeSurfer is optimised for adult brain anatomy, which could lead to slightly wrong segmentation of the brain regions. However, each segmentation was verified manually. Lastly, we excluded approximately one third of the data set, amongst other things due to poor scan quality impeding a successful segmentation by FreeSurfer. This could lead to excluding scans with large amounts of motion and thus, underestimating the true motion of children in general. However, our comparison of included and excluded scans in Supplementary Figure 2 confirmed that excluded scans did not systematically have larger amounts of motion, apart from two outlier scans (see Supplementary Figure 2).

Considering the presented results together with the discussed limitations, our study puts some light onto how children move during a (PET/)MR scan, when they are awake as well as anaesthetised. First of all, we showed that even children under GA show an amount of motion on the order of 1–2 mm mean displacement, which can impact MR image quality. This strengthens the need for alternative methods such as motion correction techniques and adequate preparation of the children to avoid motion artefacts in case the motion causes clinically image degrading artefacts. For the anaesthetised children, the clinical evaluation of image quality reported in this study shows no problem with motion artefacts; free motion, however, requires motion correction to avoid GA for more children. In addition, our data indicates that clinical routines as well as training methods under development should be given special attention to prevent nodding motion. Similarly, the higher prevalence of nodding motion and translation along the z-axis should be taken into account when optimising and testing motion correction methods. In order to limit motion artefacts further, the application of MR sequences that are more robust toward these two types of motion regarding slicing and phase encoding direction should be considered.

The data analysed in this study is subject to the following licenses/restrictions: due to regulations according to the Danish Data Protection Agency, the very sensitive patient data including the MRI sequences are not available. Requests to access these datasets should be directed to Lisbeth Marner, lisbeth.marner@rh.regionh.dk.

The studies involving human participants were reviewed and approved by Danish National Committee on Health Research Ethics. Written informed consent to participate in this study was provided by the participants' legal guardian/next of kin.

Data collection was performed by LM, SK, AE, and JS. HE, A-VV, MN, and MG contributed to the design of the work and analysed the data. The first draft of the manuscript was written by HE. All authors interpreted the data, critically reviewed the manuscript, and approved the submitted version.

This work was supported by The Danish Childhood Cancer Foundation (2014-34, 2015-48), as well as the Elsass Fonden (18-3-0147), and an unrestricted Grant from the Novo Nordisk and Novozymes Talent Program.

AE and JS, as well as MG's husband are employees at TracInnovations, Ballerup, Denmark. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.