Introduction

Sarcomas account for about 10% of all solid pediatric tumors and can be grouped in soft tissues and bone sarcomas. Soft tissue malignancies can be subdivided in two main groups: rhabdomyosarcoma and non-rhabdomyosarcoma. Rhabdomyosarcoma is the third most frequent extracranial pediatric solid tumor in children and adolescents, accounting for 5% of all pediatric cancers and for almost 50% of soft tissue sarcomas [1, 2]. Non-rhabdomyosarcomas represent around 7% of all pediatric malignancies and include different histotypes like non-osseous Ewing sarcoma, peripheral nerve sheath tumors, and synovial sarcoma [1, 3]. The most common malignant bone tumor in childhood is the osteosarcoma, with a prevalence of 3–5%, followed by the osseous Ewing sarcoma [1].

Radiological imaging plays a significant role in the management of sarcomas. In particular, magnetic resonance imaging (MRI) is usually applied for assessing the primary tumor, while whole-body computed tomography (CT) is used for staging. Over the past few decades, the application of hybrid imaging has progressively increased since it allows the simultaneous collection of anatomical and metabolic information improving tumors’ characterization. In particular, recently, there is growing interest on positron emission tomography (PET)/MRI, which carries several advantages in comparison with PET/CT [4, 5]. In fact, PET/MRI combines local and whole-body staging allowing also the use of peculiar sequences like diffusion weighted imaging (DWI) and MR spectroscopy [5,6,7,8,9,10].

Despite the growing body of evidence about DWI and PET/MRI, up to now, studies investigating the relationship between SUV and ADC using a volumetric approach in pediatric patients with sarcoma were still missing. Thus, aim of our project was to assess this relationship using simultaneously collected volumetric data extracted from pediatric sarcomas at staging, examined by PET/MRI.

Materials and methods

Patients

All patients’ parents gave written informed consent before imaging. This retrospective study was approved by the Institutional Review Board.

Pediatric patients with histologically proven sarcoma (according to the World Health Organization Classification) referring to our tertiary center from January 2015 to July 2018 who underwent a whole-body 18F-FDG PET/MRI examination for staging, including DWI of the primary lesion in the protocol, were included.

PET/MRI protocol

Patients fasted 4–6 h before the examination, keeping a proper hydration. Glucose hematic concentration was assessed, with a desirable value < 120 mg/dl (6.66 mmol/l). The dose of radiotracer was based on patient weight (3 MBq/kg). Patients were kept warm and at rest for 45–50 min after the 18F-FDG injection. Sedation was performed in case of very young and less compliant patients.

Whole-body PET/MRI images were acquired with an integrated 3 T PET/MRI scanner (Biograph mMR, Siemens Healthcare, Germany). The PET/MR system consisted of a rigid 16-channel head/neck coil, a rigid 24-channel spine array coil, up to four flexible six-channel body matrix radiofrequency coils, an integrated PET scanner, installed between the gradient and radiofrequency coils, consisting of 8 rings of 56 detector blocks, 8 × 8 lutetium oxyorthosilicate crystals per block (4 × 4 × 20 mm), coupled to an array of 3 × 3 avalanche photodiodes. The axial range of each bed position was 25.8 cm with a 6.1-cm overlap between the adjacent bed positions.

Whole-body Dixon volumetric interpolated breath-hold examination (VIBE) MR images were used as attenuation correction maps. PET data were reconstructed by the 3D-ordered subset expectation maximization algorithm (3D-OSEM). A field of view (FOV) of 400 mm and a slice thickness of 3 mm have been applied. The MR protocol included the following whole-body sequences: axial T1-weighted (T1w) turbo spin echo, axial T2w half-Fourier-acquired single-shot turbo spin echo (HASTE), axial turbo inversion recovery magnitude (TIRM) (Table 1). A single-shot echo planar imaging sequence was used for DWI (b = 50/1000 s/mm2). Fat-saturated VIBE images of the primary lesion were acquired before and after intravenous injection of gadolinium (gadoterate meglumine, 0.2 ml/kg). The total scan time was of ca. 50–60 min.

Table 1 Parameters of the magnetic resonance sequences included in the applied PET/MRI protocol
Full size table

Images analyses

PET and MRI images were evaluated in consensus by one radiologist and one nuclear medicine physician respectively with five and fifteen years of experience in pediatric imaging. For each patient, only the primary tumor was considered for the analyses. All volumetric analyses were performed by an open-source software (OsiriX, v.8, https://www.osirix-viewer.com) applying multiple steps. Initially, to resize the datasets to the same dimensions, spacing, and origin, reslicing and resampling of PET, TIRM, and ADC images were performed. Then, a volumetric manual segmentation along the border of the primary tumor and throughout all slices were performed using the TIRM dataset. Each volumetric ROI was then applied on the PET and ADC datasets. Pixel-based SUV and ADC values of the entire volume were then collected (Fig. 1).

Fig. 1
figure 1

Representation of the methodology applied to extract the pixels-based ADC and SUV values from the entire volumetric images, of each lesion, obtained by a fully integrated PET/MRI scanner. In particular, it is hereby represented a 10-year-old male with a rhabdomyosarcoma of the right quadriceps muscle characterized by high FDG uptake and restricted diffusion. A region of interest (ROI) was drawn on the turbo inversion recovery magnitude images on the axial plane (blue dotted line) covering the entire volume and then after reslicing and resampling, the same ROIs were copied on the ADC map (same blue dotted line in b) and PET dataset (same blue dotted line in c). All pixels were then extracted and analyzed. In c and d are represented, respectively, the histogram of the pixels obtained from the ADC map and the 3D volumetric reconstruction of the lesion of the same patient

Full size image

Statistical analysis

Descriptive statistics were applied for categorical data. Mean, median, skewness, and kurtosis of SUV and ADC values were computed. To evaluate the relationship between metabolic activity and cell-density, the Pearson correlation coefficient was applied for the entire population and, separately, for histological subgroups including more than five patients.

All statistical analyses were performed with SPSS (SPSS Statistics vr 24.0, IBM Corp, Armonk, NY).

Results

Thirteen patients met the inclusion criteria (seven males; mean age ± standard deviation (SD) 8.31 ± 6.03 years). Nine children were affected by rhabdomyosarcoma (i.e., five alveolar and four embryonal), three by osseous Ewing sarcoma and one by chondroblastic osteosarcoma (Fig. 2).

Fig. 2
figure 2

Axial TIRM in (a) well demonstrating the rhabdomyosarcoma of the right biceps muscle affecting a 11-year-old female. The lesion demonstrated restricted diffusion (b) and high metabolic activity (c)

Full size image

Nine patients had a localized disease, while four had distant metastases at diagnosis. Five primary lesions were in the leg, four in the hip/pelvis, two each in the head/neck and in the arm. The mean ± SD volume of the primary lesions at staging was of 114.73 ± 154.89 cm3.

The distribution of mean, median, skewness, and kurtosis of the ADC and SUV values is represented in Fig. 3.

Fig. 3
figure 3

Box plots representing the distribution of mean (a, b), median (c, d), skewness (e, f), and kurtosis (g, h) of the ADC and SUV values deriving from the volumetric analyses

Full size image

Regarding the volumetric analyses, considering the entire population, a moderate to good significant negative correlation occurred for ADCs’ and SUVs’ mean (rmean = − 0.501, p < 0.001), ADC and SUV median (rmedian = − 0.519, p < 0.001), and ADCs’ and SUVs’ skewness (rskewness = − 0.550, p < 0.001) (Fig. 4). A poor significant positive correlation between ADCs’ and SUVs’ kurtosis (rkurtosis = 0.346, p < 0.001) was found (Fig. 4).

Fig. 4
figure 4

Scattered plots representing the relationship between ADC’s and SUV’s mean (a), median (b), skewness (c), and kurtosis (d)

Full size image

Since nine cases of rhabdomyosarcoma were collected, separate analyses for this group were then performed. A moderate to good significant negative correlation between ADCs’ and SUVs’ mean, median, and skewness (rmean = − 0.541, rmedian = − 0.597, rskewness = − 0.568, p < 0.001) occurred, whereas a poor significant positive correlation between ADCs’ and SUVs’ kurtosis (rkurtosis = 0.348, p < 0.001) was detected.

Discussion

Our results demonstrated that there is a significant correlation between volumetric ADC and SUV values in pediatric sarcomas examined for staging. In particular, mean and median values showed a significant reciprocal correlation. This evidence is consistent with the fact that malignancies are characterized by low ADC values due to high cellularity and high SUV values caused by the high metabolic rate. Our data also in line with previous studies on adults [9]. For instance, Lahiji and colleagues showed that the extent of restricted diffusion and overlap hypermetabolism in patients with rhabdomyosarcoma have a negative predictive role [11]. Moreover, Sagiyama et al. demonstrated a negative correlation between voxel-based ADCs and SUVs values, collected by PET/MRI, which was even stronger for high grade tumors [11]. Similar results were obtained by a PET/CT and MRI study on skeletal and soft tissue tumors, using ADCmin and SUVmax values collected from a single slice ROI [12]. Such relationship emerged also for head and neck squamous carcinomas, rectal adenocarcinomas, and primary cervical cancer [8, 13,14,15]. On the contrary, controversial results emerged for other tumors like breast cancer or lymphomas [7, 16].

Several factors may have contributed to this heterogeneity. Firstly, some studies were performed with PET/CT and MRI [8, 13,14,15,16,17,18] without a simultaneous acquisition of data. This aspect may have influenced the measurements not only in terms of biological modifications of the lesions, but also regarding the fusion of images collected from different devices.

Secondly, most of the studies were conducted using 2D ROIs [8, 9, 19] and a single ROI approach may not properly represent the heterogeneity of the tumor. It is especially important for sarcomas because they are often characterized by necrotic and/or hemorrhagic areas, collagen fibers and mucinous fluid in addition to tumor cells [9]. Thus, our volumetric analyses provided a better insight into tumors’ structure.

Among the volumetric histogram analyses that can be collected, mean, median, and standard deviation were mostly investigated in the literature. Skewness, which represents a measure of asymmetry of the probability distribution and kurtosis that reflects the tails of the distribution were assessed in a few studies [20,21,22]. For example, they correlated with the histological grade in endometrial cancer and were good predictors of the response to neoadjuvant chemotherapy in locally advanced rectal cancer [23, 24]. In other malignant tumors, higher skewness, and lower kurtosis were associated with higher histological grades and advanced diseases’ stages [8, 20].

In our study, the significant negative correlation between the skewness of ADC and SUV values, also in the subgroup of rhabdomyosarcomas, demonstrated, as mentioned above, an overall high metabolic rate of lesions with high cellularity despite the known heterogeneous behavior of sarcomas at both, PET and at DWI.

The poor positive correlation of kurtosis does not allow us to draw precise conclusions about its role. Nevertheless, as previously shown, this variable could contribute to the evaluation of diseases’ progression. Therefore, future studies focusing on restaging are highly encouraged [25].

Furthermore, no significant correlation was found between ADCs’ and SUVs’ standard deviation. This evidence might be due to the high heterogeneity of signals on the ADC map due to different components of the tumor [11]. However, a correlation between histological and radiological tumors’ architecture is necessary to fully assess this evidence.

It should also be noted that neither ADCs’ nor SUVs’ volumetric histograms correlated with tumors’ volume. This result is in agreement with the existing literature about this type of cancer. In fact, a negative correlation between tumor volume and ADC values has been reported for several tumors but sarcomas [26].

Overall, our preliminary results support the application of texture analysis in pediatric musculoskeletal imaging for staging and we call for further studies investigating its prognostic role. Moreover, we comprehensively promote the role of PET/MRI. In fact, this technique delivers a lower radiation dose, which is a key element in oncological patients undergoing long-term follow-up [27]. The use of PET/MRI may reduce the total number of examinations since it combines local and whole-body staging.

Our study is affected by several limitations. Firstly, we examined a small sample size, albeit pediatric sarcomas are rare tumors. Despite the low number of cases, we have also performed a subgroup analysis for rhabdomyosarcomas, that is the most common soft tissue sarcoma of childhood. Therefore, we can assume that our results are representative for this type of cancer.

Secondly, we did not directly correlate radiological data with histological findings, but not all patients underwent surgical resection. Future studies focused on pathological and radiological correlation may provide further insights.

Moreover, we did not perform any longitudinal analysis after treatment. Surely investigating the relationship between voxel-based SUVs and ADCs at follow-up, considering also the long-term outcome and the eventual prognostic capacity, might be of interest for the scientific community.

Our population was mainly composed of typically FDG-avid tumors, but it should not be overlooked that sarcomas are not always hypermetabolic. Thus, our results may not be considered representative for all sarcomas.

Lastly, we have used a manual segmentation which is time consuming. We call for studies, taking advantage of methods of automatic segmentation which may also promote the application of this type of analyses in the clinical practice [28].

Future perspective

The field of quantitative musculoskeletal oncological imaging is expected to further expand in the near future, taking full advantage of the most recent technical developments. For instance, the relationship between innovative/specific tracers and DWI should be investigated. In fact, several studies obtained interesting results using tracers allowing the noninvasive assessment of hypoxia such as [18F]HX4 in a preclinical rhabdomyosarcoma model, 18F-FMISO for brain and prostatic tumors, and 64/60Cu-ATSM, for instance, for cervical cancer and soft tissue sarcomas. Nevertheless, PET/MRI applications including complex quantitative evaluations are still missing [29,30,31].

Additionally, a natural progression of our work will be to evaluate the role of radiomic features collected from hybrid techniques in pediatric sarcomas. Indeed, as demonstrated for other types of tumors, this quantitative approach may expand our knowledge about tumors characteristics [32, 33].

Last, the application of PET/MRI is also expected to explore the benefit of additional techniques like tumors perfusion, MR spectroscopy, intravoxel incoherent motion in association with the simultaneous assessment of tumors’ metabolic activity [34, 35].

In conclusion, according to our preliminary PET/MRI study, volumetric simultaneously collected SUVs and ADCs are dependent biomarkers in FDG-avid sarcomas. Further studies including the response to treatment are necessary to fully assess the clinical impact of this evidence.