Hybrid PET/MRI in Staging Endometrial Cancer: Diagnostic and Predictive Value in a Prospective Cohort

Purpose. Assessment of deep myometrial invasion (MI) and lymphovascular space invasion (LVSI) is of utmost importance in preoperative staging of endometrial cancer (EC). Imaging parameters derived respectively from MRI and PET have shown good predictive value. Aim of the present study is to assess the diagnostic performance of hybrid 18F-FDG PET/MRI in EC staging, with particular focus on MI and LVSI detection. Methods. Prospective monocentric study including 35 patients with biopsy-proven EC undergoing preoperative 18F-FDG PET/MRI for staging purpose. Histological examination was the reference standard. PET (SUVmax, SUVmean40, MTV40, TLG40) and MRI (volume index-VI, total tumor volume-TTV, tumor volume ratio-TVR, ADCmean, ADCmin) parameters were calculated on the primary tumor, and their role in predicting histological ndings (grade, high- vs. low-risk groups, LVSI, MI, p53 hyper-expression) was assessed through a ROC analysis. accuracy, and 27 risk TLG40 LVSI to predict grading, MI nor p53 hyper-expression.

Speci c intrinsic disease characteristics also in uence outcome. In stage I patients, elevated myometrial invasion (> 50%) and tumor grade (III) play a further discriminating factor on prognosis and increases the risk of pelvic and distant recurrence with a signi cant difference in 5-year survival which is reduced to 58% [8][9][10]. The 5-year OS of patients with non-endometrioid serous tumors (20-25%), usually diagnosed at later stages in which extrauterine disease is present in 60-70% of cases, is poor [7,[11][12][13].
Surgical intervention is commonly planned based on information provided by preoperative biopsy (histotype, tumor grade) and morphological imaging derived data (such as myometrial invasion and lymph node involvement). However, basal histotype may be elusive due to intrinsic tumor heterogeneity which may not be represented in the biopsy sampled tissue [14,15]. Current guidelines also indicate the best adjuvant therapies based on precisely de ned risk groups which derive from histopathology [6,7].
Magnetic resonance imaging (MRI) provides high soft tissue contrast and is particularly useful in local staging of the disease. MRI can accurately assess the depth of myometrial invasion which can impact on patient management and aid in stratifying patients into low versus intermediate risk groups before surgery [6,[16][17][18].
Evaluation of deep myometrial invasion and lymph node involvement derived respectively from MRI and PET studies in patients with endometrial cancer has shown to have good predictive value [21,22].
In such a complex scenario, the use of a fully integrated PET/MRI scanners, which combines high contrast resolution of soft tissue from MRI with metabolic information deriving from the PET component, is an undoubtfully advantageous providing an accurate local staging and whole-body staging of the disease.
The aim of the present study is to evaluate the performance of hybrid 18F-FDG PET/MRI in staging endometrial cancer patients. In particular, the diagnostic accuracy of this modality in detecting the depth of myometrial invasion and the presence of lymph node metastases in patients with endometrial cancer candidate to surgery will be investigated. In addition, the prognostic value of speci c PET and MRI derived parameters in detecting the risk of disease recurrence and disease aggressiveness will also be assessed.

Patients
This is a single center prospective cohort study carried out at San Raffaele Scienti c Institute.
All consecutive female patients presenting to the Hospital's Department of Obstetrics and Gynecology with histopathological con rmation of primary endometrial cancer, scheduled to 18F-FDG PET/MRI and candidate to surgical intervention have been enrolled in the study between December 2018 and January 2021. Data was right censored on March 31st, 2021. Informed consent was obtained from all participants prior to inclusion. Clinical history, demographic data and histology were collected. Follow-up data collected until March 2021 included imaging, clinical and laboratory data, need of adjuvant therapy (chemotherapy and/or external beam radiation therapy).
A dichotomous classi cation of risk was used to de ne low-and high-risk groups according to the ESMO-ESGO-ESTRO consensus conference [6]: patients with type 1, stage 1 and grade 1-2 and negative for LVSI were regarded as low-risk; the remaining patients were included in the high-risk group. Deep myometrial invasion was de ned as myometrial invasion > 50%.
The study was approved by the Institution's Ethics Committee (protocol number 85/INT/2019). All procedures were carried out in accordance with the Declaration of Helsinki (1964) and its later amendments; informed consent was collected from all patients according to EC guidelines.
PET/MR Imaging acquisition All scans were performed using a fully hybrid 3T PET/MRI system (SIGNA PET, General Electric Medical Systems, Wakesha, Wisconsin, USA).
A simultaneous PET/MRI Whole Body (WB) scan was acquired approximately 60 minutes (range 60-90) after i.v. administration of 18F labelled deoxy-glucose (18F-FDG).The dose injected was calculated following the EANM procedure guidelines for tumor imaging [23]. The acquisition protocol started with MR localizers, in order to de ne the number of table positions (PET-FOV) needed to cover the spatial range for the WB study. PET acquisition was performed in the 3D list mode (4min/PET-FOV) to ensure extensive data handling during image reconstruction. Simultaneously, a two-Point Dixon MR sequence was acquired (for each PET-FOV) for the attenuation correction of the corresponding PET data. LAVA-Flex (LAVA: liver acquisition with volume acceleration) sequences were also acquired over the WB scan range and used for anatomical localization of the PET signal.
At the end of the WB scan and after a brief break, a dedicated single PET-FOV centered on the pelvic region was acquired. Prior to acquisition, patients were administered 40 mg of hyoscine butyl bromide A dynamic contrast enhanced (DCE)-MRI study of the primary tumor was performed after administration of 0.1 mmol/kg of body weight of gadolinium chelate (Gadovist; Bayer, Canada) by using a T1w Spoiled Gradient Recolled Echo (SPGR) DIfferential Sub-sampling with Cartesian Ordering (DISCO) sequence (repetition time msec/echo time msec, 5.167/1.674; matrix, 148x132; number of signals acquired, 1.42; FOV, 29x29; section thickness, 3 mm) with a temporal resolution of 4 sec. Sequential images were obtained before administration of contrast medium to 5 minutes after. Finally, a small FOV highresolution 3D fat-saturated T1-weighted sequence of the uterus (repetition time msec/echo time msec, 9/1.3; matrix, 268x268; number of signals acquired, 1; FOV, 28; section thickness, 1 mm) was acquired: this data set was subsequently reconstructed with respect to the 3 planes of the uterus in order to obtain accurate delineation of tumor boundaries.
The total acquisition time per patient was approximately 1 hour.

Image Processing and Reconstruction
From the two-Point Dixon MR sequence the resulting images (in-phase, opposed-phase, water-weighted and fat-weighted) were processed for attenuation correction (AC) to generate a four-class segmented map of attenuation coe cients (air, lung, water, fat) used for AC of the corresponding PET data.
PET image reconstruction was performed using a Bayesian penalized likelihood algorithm (Q.Clear-GE) which allows effective convergence of image accuracy while suppressing noise, using a penalty function.
The algorithm includes Point Spread Function and Time of Flight models. Twenty-eight subsets and twenty-ve iterations were performed to reach signal convergence. The beta parameter, which characterizes the weight of the penalization terms, was independently optimized for Whole Body and high statistics (single FOV) PET scan. Image analysis PET images were analyzed by an experienced nuclear medicine physician (F.F. 20 years' experience) and MRI images by an experienced radiologist (G.I. 7 years' experience), both blinded to clinical and instrumental data. A consensus was reached at the end of the evaluations for the production of a combined PET/MRI evaluation and report. Qualitative and quantitative analysis was performed with the Advantage Workstation (ADW Dexus Version 4.7) of both PET and MRI derived data.
Regarding PET image analysis, the presence of pathological 18F-FDG uptake was de ned as an increased tracer uptake compared to the tracer distribution in the surrounding tissues and structures, with the exception of areas of physiologically increased uptake. Exact anatomic location of abnormal uptake in the uterus and lymph nodes was identi ed on the basis of integrated MRI images; lymphnodes were considered positive if 18F-FDG uptake was evident, regardless of their diameter on MR images.
Quantitative analysis of the primary tumor on trans-axial PET images included 3-dimensional volumetric measurements of: maximum standardized uptake value (SUVmax), mean standardized uptake value (SUVmean), Metabolic Tumor Volume (MTV) and Total Lesion Glycolysis (TLG). SUVmax was de ned as the maximum activity concentration in the tumor/(injected dose/body weight), MTV as the FDG-avid tumor volume, and TLG as MTV × SUVmean, where SUVmean was de ned as the mean concentration of FDG in the tumor/(injected dose/body weight). MTV was obtained as the sum in cubic centimetres of the tumor volume on the FDG PET scan. For SUVmean, MTV and TLG the threshold was set at 40%.
MRI interpretation followed classi cation of the primary lesion (multiparametric approach combining T2weighted images, DWI and T1-weighted post-contrast images) according to current guidelines [17,24]. Pelvic lymph nodes with a short axis diameter greater than 8 mm were classi ed as malignant [6]. In addition, descriptive morphological criteria (round shape, speculated margins, heterogeneous signal intensity or the presence of internal necrosis) were used to describe other node involvement.
The evaluated quantitative MRI parameters were: volume index (de ned as products of maximum anterior-posterior, transverse and cranial-caudal tumor diameters), total tumor volume (TTV), tumor volume ratio (TVR), ADC (apparent diffusion coe cient) -mean and -min. TVR is de ned as (TTV/TUV) x 100; TTV is the total tumor volume measured on fat-saturated T1-w post-contrast images (better delineation of lesion boundaries); TUV is total uterine volume measured on high-resolution T2w sagittal images (excluding the cervix and broids deforming the outer edge of the uterus). For apparent diffusion coe cient (ADC) calculation, ROIs (regions of interest) were drawn around the tumor on each section on the high b-values images and then these ROIs were copied on the ADC map. Perfusion evaluation and analysis is not described in the present study.

Surgery and histopathological analysis
All patients underwent surgical intervention within 1 month from the 18F-FDG PET/MRI scan. Surgery consisted in total open or laparoscopic hysterectomy, bilateral salpingo-oophorectomy and peritoneal washing. A pathologist specialized in gynecologic oncology (G.T. with more than 30 years of experience), blinded to PET/MRI ndings, performed histopathologic examination of all cases with multiple sections for each case. For each case, histotype, grading, level and type of myometrial in ltration pattern, lymphovascular space invasion (LVSI) were evaluated. In addition, a panel of Immuno-histochemical parameters were also considered: estrogen receptor (ER), progesterone receptor (PGR), p16, p53, Beta-catenin and Micro-Satellite Instability (M.S.I). Positivity for p53 was correlated to wild type or mutational type expression (hyper-expressed or null).
For nodal staging, histopathological ndings after pelvic/paraortic lymphadenectomy or sentinel lymph node dissection (SLND) as well as imaging follow-up were used as a reference standard.
Staging was assessed according to the FIGO classi cation of endometrial tumors.

Statistical analysis
In order to investigate the prognostic role of the fully integrated PET/MRI exam, several parameters extrapolated from the PET and MR images were evaluated with respect to their ability in predicting histological ndings after surgery (risk group, deep myometrial invasion, grade, LVSI, p53 hyperexpression) through a Receiver Operating Characteristics (ROC) curve analysis. For each PET/MRI parameter, the optimal cut-off was derived using the standard method, consisting in choosing that value corresponding to the point on the ROC curve nearest to the upper left corner of the ROC graph. For each histological variable, adjustment for multiple comparisons was performed using Bonferroni's correction.
Overall, PET/MRI ability (as a qualitative variable) to detect lymph node involvement and deep myometrial invasion was evaluated with the following indices: sensitivity, speci city, accuracy, positive (PPV) and negative (NPV) predictive value.
P-values less than 0.05 were considered signi cant. All statistical analyses were performed using R 3.5.0 (http://www.R-project.org/).

Patients' characteristics
A total of 36 women were enrolled in the study, yet only thirty-ve women were included in the nal analysis as one patient did not follow-through with planned surgery (extensive metastatic disease). Criteria for selection are depicted in Fig.1. Mean age was 66.57 years (SD: 10.21). No adverse events were recorded during the acquisition of the PET/MRI scan and none were reported until end of data collection.
Six patients did not undergo SNL biopsy, sample or lymphadenectomy and were evaluated at de ned imaging and clinical time-points (minimum follow-up of these patients 1-year post-surgery). Of these, only one presented lymph node metastasis at 9 months post-surgery imaging. The remaining patients underwent pelvic systematic lymphadenectomy (6), SLND (13) or biopsy sampling (10): in 6 patients' nodal involvement was positive. Therefore, a total of 7 women (20%) had lymph node metastases.
Complete patients' demographics and tumor characteristics based on histopathological data (including Grade and Figo score) are presented in Table 1.

18F-FDG PET/MR image ndings
Focal pathological 18F-FDG uptake was detected in correspondence to the primary tumor in all patients (100%). In 8 patients (28%), increased focal uptake was also detected in correspondence of pathological lymph nodes (patient-based).  Fig.2 and Fig.3). Complete imaging-based parameters are reported in Table 2. 18F-FDG PET/MRI performance and predictive value PET/MRI was able to detect lymphnodal involvement with high accuracy and high speci city (sensitivity=0.8571, speci city=0.9286, accuracy=0.9143). Hybrid imaging also had a high negative predictive value for lymphnodal involvement (NPV=0.9630; PPV=0.7500).
ROC curve analysis used to investigate the performance of selected PET and MRI quantitative parameters in predicting tumor status or characteristic (derived from histology) well described the prognostic role of PET/MRI. MRI derived volume index, TTV, and TVR were signi cant in predicting endometrial cancer groups (high risk vs. low-risk patients) (p= 0.0059, 0.0235, 0.0181, respectively; Table 3a). MRI derived volume index, TTV, TVR and PET derived MTV40 and TLG40 were able to predict LVSI (p=0.0023, 0.0068, 0.0068, 0.0027, 0,01394, respectively; Table 3b). Imaging was not able to predict grading, presence of deep myometrial invasion, hyper-expression of p53 (Table 3c-e). ROC curve analysis for positive lymph nodes prediction was not performed due to the small number of events.

Discussion
Imaging has a central and undisputed role in the preoperative evaluation of patients with endometrial cancer [25-27] and the ESMO-ESGO-ESTRO consensus conference highlighted its relevance for surgery planning, especially when considering lymphadenectomy and the associated differences in terms of risk of recurrence and prognosis [6]. Imaging derived biomarkers are often used in clinical and research settings, as objective, non-invasive indicators of biological, pathological and prognostic factors. The most widely used in EC are the SUV for PET imaging and ADC for MRI. In particular, SUVmax is correlated with tumor aggressiveness and prognosis, while ADCmin is useful for outcome prediction [28-31].
In our patient population, hybrid 18F-FDG PET/MRI correctly detected all primary tumors and had a good accuracy in detecting myometrial invasion (77%) and a high accuracy in detecting lymph node involvement (91%). Volume MRI imaging derived parameters (Volume index, TTV, TVR) and PET-derived MTV40 and TLG40 can predict LVSI and MRI volume parameters can correctly stratify patients with highversus low-risk pro les before surgery.
These quantitative evaluations could play a decisive role in planning therapeutic and surgical choice in view of personalized treatment approaches.
Previous studies have demonstrated a relevant role of the combined use of both 18F-FDG PET and MRI in the staging setting of endometrial cancer. For instance, our group demonstrated a synergic ability of these two imaging modalities, used separately, in providing an accurate staging of EC, also identifying correlations between PET and MRI derived parameters [32-34].
However, very few studies have investigated the accuracy and predictive value of simultaneous 18F-FDG PET/MRI in the staging setting of endometrial cancer, so far. Volume index, TLG and SUVmax/ADCmean ratio derived from distinct MRI and PET scans predict myometrial invasion and lymph node involvement or disease recurrence [21,27,[32][33][34].
Only two studies [35,36] [36] retrospectively reported a sensitivity of 100% for 18F-FDG PET/MRI in detecting the primary tumor and an accuracy of 81.8% in delineating myometrial invasion; the slightly higher percentage of detection compared to our results may be related to the retrospective design of the study [36].
The predictive role of tumor size in patients with endometrial cancer has been also previously demonstrated. The FIGO staging system considers tumor size as one of the standard measurements for cervical cancer, but not for endometrial cancer [8] even though its prognostic value can be traced back to the second half of the last century, when Gusberg et al. [37] demonstrated positive correlation between tumor size and poor prognosis. Schink et al. [38] used average tumor diameter to predict lymph node involvement and survival; tumors > 2 cm were correlated to higher rate of lymph node metastasis (15% vs 4% ≤ 2 cm) and lower survival (98% 5-year survival for patients with tumors ≤ 2; 84% 5-year survival > 2 cm). Mariani et al. [39] reported that there was a signi cant difference in prognosis for patients with lowrisk endometrial cancer and tumor size (< 2 cm and ≥ 2 cm). Myometrial invasion, histology and tumor size are also used in the Mayo criteria for risk assessment [40].
Our data shows that volume index (> 15.29) is associated with high-risk tumor and a volume index > 25.45 is also predictive for LVSI. This is coherent with what reported in literature where high volume index is associated to occurrence of lymph node metastases [41] and poor prognosis [42]. The ratio between tumor and uterine volume (TVR) > 21.1 was associated with lymphovascular space invasion; a value very similar to that found by Nougaret et al. [43] in which the cut-off was ≥ 25.36.
In our sample, no association was observed between MRI diffusion parameters and histology (grade, LVSI, deep myometrial invasion, risk or p53 expression), although this result could be due to the limited sample size. Similarly, Rechichi et al. [44] described that ADC had no correlation to grade and myometrial invasion. Shih et al. [34] described that ADCmin extrapolated from a fully integrated PET/MRI scan was associated to grade, stage and cervical invasion, but also these authors did not nd a correlation with myometrial invasion and LVSI. Bharwani et al. [45] showed a non-statistically signi cant difference between tumor grades and ADCmin-and mean. ADC values were correlated to tumor grade by Tamai et al. [46] however, the authors described a certain overlap of ADC ranges for the different grades. Therefore, the correlation between ADC and tumor grade is still a matter of discussion.
Regarding risk groups, Tsuyoshi et al. [35] reported that high-risk cancers had a signi cantly lower ADC than low-risk cancers. These results are in contrast with those of our study, in which no association was found between these parameters. Further studies with larger patient numbers are needed to clarify the discording role of ADC in endometrial cancer.
Even though the most widely imaging biomarker in PET is SUVmax, usually associated with tumor aggressiveness and prognosis [28, 33, 34], this nding is not con rmed by our data with the actual sample size. However, we found a predictive value of MTV40 and TLG40 in assessing LVSI. An interesting and additive value of our study relies on the fact that we provided speci c threshold of these metabolic parameters, able to indicate the presence of a higher risk of aggressiveness, as it is LVSI. No PET derived parameters were able to differentiate low-versus high-risk patients.
The limitation of the data presented may be the low number of our patients' cohort, yet this is only a preliminary analysis of a larger prospective population of endometrial cancer patients in which PET/MR images are simultaneously acquired. In addition, the low number of patients with lymph node involvement did not allow to well perform a statistical evaluation of the ability of hybrid imaging derived parameters in predicting this event.
In conclusion, fully integrated PET/MRI shows good to high accuracy in detecting myometrial invasion and lymph node involvement. Moreover, distinct quantitative parameters from this hybrid modality of imaging can predict the presence of LVSI and can correctly stratify patients according to the internationally accepted risk pro les.

Declarations
Compliance with ethical standards

Con ict of interest
The authors declare no con ict of interest.   Figure 1 Flow diagram of our prospective single-Centre cohort study