Perfusion imaging of neuroblastoma and nephroblastoma in a paediatric population using pseudo-continuous arterial spin-labelling magnetic resonance imaging

Objectives To examine the feasibility of performing ASL-MRI in paediatric patients with solid abdominal tumours. Methods Multi-delay ASL data sets were acquired in ten paediatric patients diagnosed with either a neuroblastoma (n = 4) or nephroblastoma (n = 6) during a diagnostic MRI examination at a single visit (n = 4 at initial staging, n = 2 neuroblastoma and n = 2 nephroblastoma patients; n = 6 during follow-up, n = 2 neuroblastoma and n = 4 nephroblastoma patients). Visual evaluation and region-of-interest (ROI) analyses were performed on the processed perfusion-weighted images to assess ASL perfusion signal dynamics in the whole tumour, contralateral kidney, and tumour sub-regions with/without contrast enhancement. Results The majority of the included abdominal tumours presented with relatively low perfusion-weighted signal (PWS), especially compared with the highly perfused kidneys. Within the tumours, regions with high PWS were observed which, at short PLD, are possibly related to labelled blood inside vessels and at long PLD, reflect labelled blood accumulating inside tumour tissue over time. Conversely, comparison of ASL perfusion-weighted image findings with T1w enhancement after contrast administration showed that regions lacking contrast enhancement also were void of PWS. Discussion This pilot study demonstrates the feasibility of utilizing ASL-MRI in paediatric patients with solid abdominal tumours and provides a basis for further research on non-invasive perfusion measurements in this study population.


Introduction
In paediatric oncology, imaging plays a defining role in the initial diagnosis, staging, early post-treatment response assessment and surveillance. Magnetic resonance imaging (MRI) is currently the preferred modality for clinical assessment of tumours located in the abdomen [1]. Clinicians rely mainly on anatomical imaging techniques for staging, risk stratification and assessing treatment response [2]. The state-of-the-art method to evaluate response to treatment in solid tumours is by measuring the tumour size following the RECIST (Response Evaluation Criteria in Solid Tumours) guidelines [3][4][5]. In addition to anatomic assessment, functional measures such as perfusion and diffusion may provide complementary information for the clinical assessment of tumour diagnosis, staging, and treatment response [4,[6][7][8][9].
The characterisation of tumour hemodynamics in children is still limited. One explanation is that childhood cancers are very heterogeneous with often relatively small patient numbers, limiting standardisation for imaging modality choice and subsequent technical optimisation [1].
Arterial spin-labelling (ASL) MRI may be a valuable method in a paediatric population for non-invasive measurement of tissue perfusion without the need for contrast administration. This method utilises magnetically labelled 1 3 water-protons in blood as an endogenous tracer. This allows for repeated measurements within the same scan session or frequent follow-up MRI examinations. Thus far, in paediatric patients ASL-MRI has been performed in a few studies for perfusion imaging of brain tumours [10][11][12][13][14]. These studies showed that ASL perfusion signal is correlated with tumour histopathological vascular density and can be used to distinguish high-and low-grade tumours. In addition, ASL renders characteristic perfusion patterns for diverse pathologic types of brain tumours in children. These findings suggest an important value of ASL-MRI in the characterisation of hemodynamics in brain tumours. This has not yet been studied in paediatric abdominal tumours.
Therefore, the purpose of this study was to examine the feasibility of performing ASL-MRI in paediatric patients with tumours located outside the brain by evaluating the presentation of ASL perfusion signal dynamics of common abdominal tumours.

Study population
This is a report on retrospective, cross-sectional data obtained in patients who received diagnostic MR imaging of the abdomen related to an abdominal solid tumour. An ASL scan had been added to the clinical scan protocol in consultation with the pediatric radiologist for the assessment of tumour perfusion. Subjects did not have to follow specific actions or rules for conducting this study, since the patients were scheduled for the MRI examination as part of the clinical diagnostic procedures. A letter waiving ethical approval was obtained from the IRB of our institution. The image data were fully anonymised before processing and analysis.

MR Imaging
All patients were scanned on a 1.5 T MR system (Ingenia, Philips, Best, the Netherlands; software release 5.4) equipped with a body coil for transmission and a 28-element phased array coil for reception.

ASL scan
The ASL scan was acquired using pseudo-continuous ASL (pCASL) labelling [15], as implemented by the vendor and clinically available as a standard sequence, with three different post-labelling delays (PLD; 0.5, 1.0, and 1.5 s) and a label duration of 1.5 s [16]. An auxiliary sequence was acquired to estimate equilibrium magnetisation (M 0 ). Each PLD was obtained in a separate acquisition that consisted of 10 label-control pairs. The M 0 scan, essentially the pCASL scan without labelling, was acquired four times and averaged after motion correction. All scans were performed with the same 2D multi-slice image readout using singleshot gradient-echo EPI. The labelling slab and image readout were planned exactly the same for each ASL acquisition. The applied scan parameters are provided in Table 1. Total acquisition time of the ASL and M 0 scans was 4min54s.
The ASL and M 0 scans were performed at the end of the MRI examination, before the injection of contrast agent. The localiser scan and subsequently acquired clinical scans were used to enable correct planning of the ASL and M 0 scans. Slices were planned in coronal orientation to minimise through-plane motion due to breathing, including as much tumour tissue as possible and (part of) the contralateral kidney within the stack thickness (62 mm). The contralateral kidney was used as a reference organ for comparison with the ASL perfusion signal measured in the tumour. Normally, the kidneys are well perfused, and therefore, measured ASL perfusion signal in this organ could be used as an indication of successful labelling within a subject. The labelling slab of the ASL scan was placed perpendicular to the descending aorta and above the tumour (and kidneys) to prevent these tissues from sliding into the labelling slab during breathing. Care was taken that the labelling slab was as close as possible to the tumour and remained below the diaphragm (if possible) to minimise susceptibility artifacts near the lungs. B 0 shimming was applied to the imaging stack and the labelling slab independently. Most patients were under anaesthesia during the entire MRI examination. Acquisitions were performed under non-invasive ventilation or free-breathing conditions in supine position.

Clinical scans
A standard clinical scan protocol was obtained in each subject, which included (among others) T 2 -weighted (T 2 w) and T 1 -weighted (T 1 w) scans. A 3D T 2 w TSE scan was acquired with TE 90 ms and isotropic spatial resolution 1.2 × 1.2 × 1.2 mm 3 . T 1 w gradient-echo imaging before and after contrast agent administration (CE-T 1 w) was performed either using the 3D THRIVE technique with fat suppression, TE 2.7 ms and acquired spatial resolution 1.07 × 1.07 × 3 mm 3 , or the 3D Vane XD technique with TE 1 / TE 2 1.9/4.0 ms (with Dixon-based water-fat separation) and acquired spatial resolution 1.5 × 1.5 × 3 mm 3 .

ASL scan
To compensate for misalignments caused by (respiratory) motion during image acquisition, image registration was performed using Elastix [17]. Since tissues and organs inside the abdominal region may move with respect to each other, each organ of interest (tumour, kidney) was registered separately. To this end, images were first cropped to the region of interest before image registration. The ASL data set was registered using a principal component-based groupwise method [18]. Multi-PLD raw ASL and M 0 images were registered slicewise (total of 64 images per slice; 4 M 0 repetitions, 10 label-control pairs per PLD). The groupwise registration method is robust against intensity changes between the images and does not require the choice of a reference image avoiding registration bias [18]. After image registration, label-control raw ASL images were pairwise subtracted. Next, outlier rejection was performed for each PLD by excluding subtraction images containing > 20% voxels with a value of more than ± 1.5 SD from the voxel wise mean value over all repetitions within the tumour or kidney region. The subtraction images remaining after outlier rejection were averaged per PLD to obtain averaged perfusion-weighted images ( ΔM ). The definition of the tumour and kidney masks that were used for the outlier rejection is described in more detail in the Image Analysis section.

Clinical scans
For assessment of contrast enhancement of the tumour, T 1 w subtraction images (T 1 w-subtracted) were used. To this end, first the CE-T 1 w scan was coregistered to the T 1 w scan using 3D rigid image registration in Elastix [17], and next both images were subtracted.

Image analysis
Regions-of-interest (ROI) analyses were performed to quantitatively evaluate the measured ASL perfusion signal in different regions for each subject. ROIs were drawn manually around the tumour and contralateral kidney contours on the averaged M 0 images (after image registration). The ROI of the contralateral kidney included the renal parenchyma. In addition, ROIs were drawn in enhancing and non-enhancing tumour subregions that were identified on T 1 w subtraction images. ROIs were drawn conservatively in the enhancing and non-enhancing tumour regions, i.e., well within the visible boundary of a non-enhancing or enhancing region, to take into account minor registration errors between the ASL perfusion and the CE-T 1 w data sets. The 3D anatomical T 2 w and T 1 w images were reformatted to the geometry of the M 0 images to cross-check the tumour contours. When necessary, ROIs were manually adjusted with in-plane translations between anatomical and ASL scans. In case multiple tumours were present, the largest tumour was used for the ROI analysis. For each tissue type (whole-tumour, contralateral kidney, enhancing tumour, non-enhancing tumour) the same ROI was used within an individual patient. ROIs were drawn by a single observer (A.H) and were verified and corrected by a radiologist (A.L with > 10 years of experience with assessing paediatric oncology MR images).
ASL perfusion signal dynamics within the tumour and contralateral kidney were assessed by calculating the relative perfusion-weighted signal (PWS) for each delay time as ΔM∕M 0 × 100% . Mean PWS values were calculated over all voxels inside the ROIs for each subject.
Image processing and analysis were performed using custom scripts in MeVisLab (version 3.2; MeVis Medical Solutions AG, Bremen, Germany).

Study population
ASL scans were obtained in 10 paediatric patients (mean age 4.3-year-old; range 1.7-8.3 years; 4 male) diagnosed with either a neuroblastoma (n = 4) or nephroblastoma (n = 6). In each patient, ASL imaging was performed at a single visit; n = 4 patients (n = 2 neuroblastoma, n = 2 nephroblastoma) at initial staging MRI examination pre-treatment, and n = 6 patients (n = 2 neuroblastoma, n = 4 nephroblastoma) at follow-up MRI-examination during treatment. Patient characteristics and primary tumour classification are presented in Table 2. Imaging, including ASL, was successfully performed in all subjects and used for further image processing and analysis.

Image processing and analysis
Image registration successfully reduced misalignment of M 0 and raw ASL images, see example, Fig. 1, enabling further perfusion analysis in all subjects. For both analysed regions (tumour and kidney), on average 1 (range 0-3) label-control pair out of 10 per pCASL acquisition (delay time) was considered an outlier and was rejected. Examples of processed images that were used for analysis are shown in Figs. 2 and 3 for a neuroblastoma and nephroblastoma case, respectively. An overview of the ROI sizes used in the analysis for each tissue type are presented in Table 3. Due to the varying sizes of the tumours that is confounded by the age range of the patients, there was a large variation in the sizes of the ROIs.

ASL PWS dynamics
A large variability in ASL PWS dynamics inside the tumour region was observed across the cases. The majority of tumours showed large regions with relatively low PWS intensity that remained stable over time for the three PLDs. Inside the tumour local regions with higher PWS intensity were observed that either increased or decreased with increasing PLD. Review of local regions with high PWS inside the tumour at the shortest PLD (0.5 s) that decreased over time indicated that these regions often coincided with local hypo-intense areas in the T 2 w images; examples are shown in Fig. 4.

Discussion
We conducted a study to establish the feasibility of ASL-MRI in paediatric neuroblastoma and nephroblastoma. The study showed that this was feasible within 5 min of extra scan time. After realignment, the acquired ASL images were usable for analysis in a paediatric population under non-invasive ventilation and free-breathing conditions. The perfusion signal in the kidney indicates that labelling was successfully performed in all subjects, yet, the majority of the included abdominal tumours presented with relatively low PWS. Within the tumours, regions with higher PWS were observed, either related to labelled blood inside vessels (at short PLD) or to labelled blood accumulating inside tumour tissue over time (increasing with PLD). Conversely, comparison of ASL perfusion-weighted image findings with T 1 w enhancement after contrast administration showed that regions lacking contrast enhancement also showed PWS around zero.
The ASL scan was obtained in a clinical setting. This defined our choice for the pCASL labelling approach, since this method was clinically available on our MR system. The pCASL labelling approach is currently recommended for brain and kidney perfusion imaging in adults [16,19]. pCASL requires the separate planning of a labelling slab at the location of the feeding arteries that must be positioned perpendicular to the blood flow direction to ensure high labelling efficiency. Since each abdominal tumour is different in size, shape, composition and location, the feeding vessels of each tumour are unique. These tumour  [20]. Ambiguity of the feeding vessel location forced us to place the labelling slab perpendicular to the descending aorta to ensure blood was labelled before entering the tumour. This location was feasible for planning of the labelling slab in all patients, but will in some patients imply placement above the diaphragm at the level of the lungs which may negatively influence labelling efficiency [21]. Tortuosity and length variation of the feeding vessels induce variability in the arterial transit time of labelled blood and the amount of label accumulation in tissue that can take place during the PLD [22]. Increasing PLD to allow time for the labelled blood to travel from the aorta into the tumour and accumulate inside the tumour tissue will limit the available PWS due to T 1 decay. Since no previous ASL studies on abdominal tumours in children have been reported yet, and hemodynamics and tumour characteristics can be very different from those in adults [14], ASL data was acquired with three different PLDs to evaluate the influence of inflow time on the perfusion signal. In the current study, high PWS was already observed at the first PLD of 0.5 s in the renal cortex, which seemed to reduce rapidly with increasing PLD (as shown in Fig. 5). This arrival and decay of renal PWS appears  Table 2). Images visualise the same intersection (yellow line in the coronal image) over all M 0 and raw ASL images. Improved alignment can be observed after image registration, for instance by looking at voxels representing the edges of the tumour/kidney  Table 2). Processed ASL and M 0 images are shown cropped around the tumour (upper row) and right kidney (bottom row), together with the reformatted T 2 w and T 1 w-subtracted images that were used for analysis of the tumour. The anatomical location of the nephroblastoma can be appreciated on the coronal T 2 w images (arrow indicates the largest tumour that was used for analysis). The graph on the right shows the perfusion-weighted to be faster than reported in adults [16], underlining the difference in hemodynamics between children and adults.
Other factors such as variability in blood and tissue T 1 values that are known to be present between children, and an age-dependency of the arterial transit time of blood flowing from the labelling plane to tissue of interest also have an influence on the measured PWS [23][24][25][26].
In general, the whole-tumour mean PWS was relatively low in all patients, especially compared with the contralateral kidney region (as shown in Fig. 5). Kidneys are normally well-perfused, so high PWS was expected in this organ. Therefore, the measured PWS values in the contralateral kidney region served as an indication that label was created and perfusion signal could be measured using this ASL method, especially since the measured PWS in the tumours was in general relatively low. The usage of large ROIs covering the entire tumour (see examples in Figs. 2 and 3) may have obscured local regions of high PWS inside the larger areas of low PWS in the calculation of the mean PWS averaged over all voxels inside this whole-tumour region. Review of local tumour regions with high PWS at the shortest PLD that decreased over time indicated that these regions often coincided with local hypo-intense areas on the T 2 w images (as shown in Fig. 4). These type of local hypo-intense areas on T 2 w images are usually caused by flow void effects inside a blood vessel. Small local areas of high PWS at the shortest PLD are, therefore, likely to reflect presence of tumour vasculature. On the other hand, a local tumour region with high PWS that increases or remains more or less constant over time may be an indication of tissue perfusion. Validation of in vivo perfusion measurements are in general difficult to perform, due to the lack of a gold standard technique.
The comparison of tumour contrast enhancement on T 1 w-subtracted images with the ASL perfusion-weighted image findings was intended to assess which aspects of tumour physiology are being measured with the applied ASL method, and whether these are physiologically plausible. In principle, the methods do not measure the same physiological processes: contrast enhancement is the result of either blood flow or vascular leakage, where ASL PWS is the result of blood flow only. Non-enhancing tumour regions (no blood flow nor leakage) should, therefore, correspond with PWS around zero (no blood flow), which was the case for all non-enhancing tumour regions in this study. Enhancing tumour regions, on the other hand, were more difficult to compare directly with ASL findings in the same tumour regions. Most tumours showed very low PWS inside enhancing regions (as shown in Fig. 6). This could either mean that the enhancement was mainly driven by vascular leakage, which usually builds up over time, or the arterial transit time of the labelled blood was too long to reach the tumour tissue and accumulate. Interestingly, there were two neuroblastoma cases with clearly higher PWS values in the  . 4 a, b Example cases from two patients (subject 2 and 9 in Table 2) showing high PWS inside the tumour on PWS images obtained with the shortest PLD of 0.5 s (PLD1) that coincided with local hypo-intense areas on the reformatted T 2 w images (white arrow heads). c This case (subject 8 in Table 2) shows very well the high PWS of labelled blood flowing inside the larger arteries (descending aorta and right renal artery; white arrow heads) indicating successful labelling. Many of the small regions with high PWS inside the tumour corresponded with hypo-intense regions on the T 2 w images, however, due to blurring effects after reformatting of the T 2 w image to the geometry of the ASL images these hypo-intense areas could not be visualised clearly here. The green contours indicate the tumour region. Colour scale bar indicates PWS (∆M/M 0 × 100%) enhancing regions, indicating a possible blood flow effect. It could also be that the arterial transit time of the labelled blood was much shorter for these two tumours, making the accumulation of PWS better visible. Thus far, ASL MRI in the body/abdomen has mainly been applied for renal perfusion imaging in adults. Recently, a consensus-based paper describing technical recommendations for clinical translation of renal ASL MRI has been published [16], that was used as a starting point for setting up the current study. For the kidneys, Echeverria-Chasco et al. [27] recently showed that the labelling efficiency and robustness of pCASL for renal applications can be optimized and depends on the aortic blood flow velocities. A similar optimization of gradient parameters for blood flow velocities in a paediatric population could be performed. However, this does not resolve the problem of heterogeneity in tumour location and feeding vessels between patients, that influence the transit-time of labelled blood from the labelling plane to the tumour tissue. This currently makes it difficult to identify the cause of low PWS inside the tumour that remains stable across all PLDs (as shown in Fig. 5) that could either be due to low or late perfusion of the tumour. It is, therefore, very important that labelling is performed as close as possible to the anatomy of interest. A flow-based labelling approach, such as velocity-selective ASL (VSASL) [28,29] and velocity-selective-inversion prepared ASL (VSI-ASL) [30], that is insensitive to transit-time and does not require the planning of a separate labelling slab, may provide a promising alternative to spatially selective ASL techniques for tumour perfusion imaging of paediatric abdominal tumours. Recently, a For clarity data points are slightly shifted with respect to each other at each post-labelling delay and one-sided error bars are shown. pCASL pseudo-continuous arterial spin labelling, PWS perfusion weighted signal few studies have already shown promising results of flowbased ASL imaging in the abdomen for the kidneys [31,32] and placenta [33,34] in adults. Another recent study in paediatric patients with Moyamoya has shown VSASL offers a powerful approach to accurately image perfusion in these patients due to transit delay insensitivity [35]. On the other hand, due to the wide availability of pCASL on clinical scanners, continuing with this labelling approach may still be a good idea, possibly after adaptation/optimization for blood flow velocities in children.
This study has limitations. A first limitation of this study is that the number of patients included was relatively small with heterogeneous characteristics, which are, however, usual limitations inherent to the nature of an explorative study on retrospective data. In general, paediatric abdominal tumours are very heterogeneous with diverse pathological features, and the number of patients for each tumour type are often relatively small. This poses challenges for conducting research with this population. So far, feasibility of ASL has been primarily studied in the brain reflecting the difficulty in evaluating new techniques in paediatric patients with rare diseases. Second, the retrospective nature of this study in a clinical setting limited the amount of time allowed for additional scans that could be performed. This meant that more extensive evaluation that would require additional scans, such as B0 and B1 maps or labelling efficiency measurements, to evaluate the technical validity of the ASL measurements or a repeatability analysis could not be performed. A   Fig. 6 ROI analysis of PWS values obtained from pCASL images within non-enhancing and enhancing subregions of the tumour in 10 paediatric patients diagnosed with neuroblastoma (black) or nephroblastoma (red). Data points represent the average PWS in non-enhancing (left) and enhancing tumour regions (right) at each delay time. A minority of two neuroblastoma cases show increased PWS in enhancing regions inside the tumour. One of the neuroblas-toma patients is missing a data point (post-labelling delay 1500 ms), since this scan was accidentally omitted during acquisition. The error bars represent the 95% confidence interval. For clarity data points are slightly shifted with respect to each other at each post-labelling delay and one-sided error bars are shown. pCASL pseudo-continuous arterial spin labelling, PWS perfusion weighted signal future study with a prospective setup would, therefore, be a logical next step. Default implemented pCASL labelling settings were used, which have been optimised for brain application in adults. Optimisation of pCASL labelling parameters for blood flow characteristics of the descending aorta in a paediatric population may improve labelling efficiency and reduce labelling variation between patients, as has been shown for renal application in adults [27]. In ASL, the labelling efficiency is mostly assumed as a constant parameter in the kinetic model for perfusion quantification [16,19,36]. In fact, in adults labelling efficiency variation may be several tens of present for settings not optimised for the abdomen [27]. In the paediatric population, which is relatively heterogeneous and with many factors that could potentially have an influence on the measured PWS, a labelling efficiency measurement on a subject basis may provide insight in the variability in this population. Third, most patients were under anaesthesia during scanning. Sedatives may have impact on tissue perfusion, complicating the interpretation of perfusion data [24,26]. The general effects of sedatives on ASL imaging are currently unknown, and remain an important factor to consider in future larger studies of paediatric imaging. Finally, due to the explorative nature of this study, ASL-MRI was only performed at a single visit in each patient, either at initial staging or during follow-up. Possible changes in tumour perfusion during treatment could, therefore, not be assessed. This should be further investigated in future studies by measuring ASL at multiple visits during treatment within the same patient.
In conclusion, this study demonstrated the feasibility of ASL-MRI with a pCASL labelling approach in paediatric patients with solid abdominal tumours. In this population considerable differences in PWS were observed both within the individual tumour and between subjects, with on average sixfold lower PWS in the whole-tumour region compared with the contralateral kidney. Within the tumour, nonenhancing tumour regions consistently corresponded with PWS around zero, whereas some enhancing tumour regions clearly showed higher mean PWS values. The preliminary data presented here provides a basis for further research on non-invasive perfusion measurements of solid abdominal tumours using ASL-MRI, specifically in paediatric patients. Further research should be directed towards improving the labelling approach, for instance by optimizing pCASL labelling settings or by exploring other labelling approaches, such as flow-based ASL techniques.