Cervical Cancer: Associations between Metabolic Parameters and Whole Lesion Histogram Analysis Derived from Simultaneous 18F-FDG-PET/MRI

Multimodal imaging has been increasingly used in oncology, especially in cervical cancer. By using a simultaneous positron emission (PET) and magnetic resonance imaging (MRI, PET/MRI) approach, PET and MRI can be obtained at the same time which minimizes motion artefacts and allows an exact imaging fusion, which is especially important in anatomically complex regions like the pelvis. The associations between functional parameters from MRI and 18F-FDG-PET reflecting different tumor aspects are complex with inconclusive results in cervical cancer. The present study correlates histogram analysis and 18F-FDG-PET parameters derived from simultaneous FDG-PET/MRI in cervical cancer. Overall, 18 female patients (age range: 32–79 years) with histopathologically confirmed squamous cell cervical carcinoma were retrospectively enrolled. All 18 patients underwent a whole-body simultaneous 18F-FDG-PET/MRI, including diffusion-weighted imaging (DWI) using b-values 0 and 1000 s/mm2. Apparent diffusion coefficient (ADC) histogram parameters included several percentiles, mean, min, max, mode, median, skewness, kurtosis, and entropy. Furthermore, mean and maximum standardized uptake values (SUVmean and SUVmax), metabolic tumor volume (MTV), and total lesion glycolysis (TLG) were estimated. No statistically significant correlations were observed between SUVmax or SUVmean and ADC histogram parameters. TLG correlated inversely with p25 (r=−0.486, P=0.041), p75 (r=−0.490, P=0.039), p90 (r=−0.513, P=0.029), ADC median (r=−0.497, P=0.036), and ADC mode (r=−0.546, P=0.019). MTV also showed significant correlations with several ADC parameters: mean (r=−0.546, P=0.019), p10 (r=−0.473, P=0.047), p25 (r=−0.569, P=0.014), p75 (r=−0.576, P=0.012), p90 (r=−0.585, P=0.011), ADC median (r=−0.577, P=0.012), and ADC mode (r=−0.597, P=0.009). ADC histogram analysis and volume-based metabolic 18F-FDG-PET parameters are related to each other in cervical cancer.


Introduction
Cervical cancer is the third most commonly diagnosed cancer and the fourth leading cause of cancer death in females worldwide [1].
Magnetic resonance imaging (MRI) has been established as the best imaging modality for staging of cervical cancers due to its excellent soft tissue contrast [2]. Furthermore, MRI can provide information regarding tumor microstructure by diffusion-weighted imaging (DWI). e principle hypothesis is that DWI can quantify the free movement of protons (Brownian molecular movement) by using apparent diffusion coefficients (ADC) [3].
is movement is hindered predominantly by cell membranes. In fact, previous studies showed that ADC inversely correlated with cell count in several malignant and benign lesions [4]. Another clinically important functional imaging modality is 18 F-fluorodeoxyglucose positron emission tomography (FDG-PET), which reflects tumor glucose-metabolism [5]. e FDG-uptake in tumor tissue is associated with the increased expression of glucose transporters (GLUT), mainly subtype GLUT-1 [6]. Clinically, 18 F-FDG-uptake is semiquantified by standardized uptake values (SUV). Moreover, it has been shown that volume-based metabolic PET parameters, such as metabolic tumor volume (MTV) and total lesion glycolysis (TLG), might provide additional information regarding tumor behavior [7]. MTV and TLG have been reported as possible prognostic factors, for example, for lung cancer or laryngeal carcinoma. In cervical cancer, for example, MTV was the only parameter to be of prognostic relevance in a multivariate analysis performed by Hong et al. [8].
Presumably, functional parameter derived from PET and from MRI, albeit reflecting slightly different tumor aspects, might be linked to each other [9]. As a hypothesis, a cell-rich tumor might also express more GLUT-transporters within their cell membranes, and hence, an association between ADC and SUV values might exist.
In fact, this was studied by various investigations in several different tumor entities like esophageal or breast cancer [9][10][11][12][13]. However, in a recent meta-analysis, comprising 35 studies, only a weak inverse correlation coefficient of r � −0.30 was identified over all various investigated tumors [9].
An emergent imaging analysis, namely, ADC histogram analysis, which is based on pixel distribution, is used to improve tumor heterogeneity in DWI-MRI assessment. Every voxel of a region of interest is issued into a histogram and thusly statistically information about the tumor is provided. Typically parameters are percentiles, median, mode, skewness, kurtosis, and entropy [17]. It is acknowledged that heterogeneity displayed by the histogram might be reflected by tumor microstructure heterogeneity, and therefore, a better reflection of tumor biology may be possible [17]. e histogram analysis approach has been applied in other tumors, for example, in prostate cancer. For example, Liu et al.
characterized histogram variables of ADC as predictors for the aggressiveness of prostate cancer [18]. In a study of Shindo et al., ADC histogram analysis has been described as helpful in differentiating pancreatic adenocarcinomas from neuroendocrine tumors [19]. Regarding cervical cancer, there are only few reports compared metabolic parameters of 18 F-FDG-PET and ADC histogram analysis. For instance, Ueno et al. evaluated the prognostic value of SUV, MTV and TLG, and ADC histogram analysis for tumor response to therapy and event-free survival in patients with cervical cancer [20]. It has been shown that pretreatment volume-based metabolic 18 F-FDG-PET parameters may have better potential than ADC histogram analysis for predicting treatment response and survival in these patients [20]. e main drawback of this study was that data from PET and MRI were obtained sequentially and not simultaneously; thus, the results of this study may have been influenced by this fact. e aim of our study was to elucidate possible associations between ADC histogram-based parameters and 18 F-FDG-PET parameters derived from simultaneous PET/MRI in cervical cancer.

Materials and Methods
is prospective study was approved by the local research ethics committee.

Patients.
Overall, 18 female patients (age range: 32-79 years; mean age: 55.4 years) with histopathologically confirmed squamous cell cervical carcinoma were enrolled. Inclusion criteria were a staging investigating with a body simultaneous 18 F-FDG-PET/MRI before any form of treatment. Table 2 gives an overview about the patients and the different clinical pathological stages.  18 F-FDG-PET/MRI (Biograph mMR-Biograph, Siemens Healthcare Sector, Erlangen, Germany) which was performed from the upper thigh to the skull for 4 minutes per bed position. PET images were reconstructed using the iterative ordered subset expectation maximization algorithm with 3 iterations and 21 subsets, a Gaussian filter with 4 mm full width at half maximum (FWHM), and a 256 × 256 image matrix. Attenuation correction of the PET data was performed using a four-tissue (fat, soft tissue, air, and background) model attenuation map, which was generated from a Dixon-Vibe MR sequence according to previous description. Radiotracer administration was performed intravenously after a fasting period of at least 6 hours with a body weight-adapted dose of 18 F-FDG (4 MBq/kg; range: 152-442 MBq; mean ± std: 285 ± 70 MBq). PET/MRI image acquisition started on average 122 minutes after 18 F-FDG application. Due to radiotracer elimination via the urinary tract, which may influence evaluation of pelvic PET images, all patients received a bladder catheter prior to PET/MRI examination.
Image analysis was performed on the dedicated workstation of Hermes Medical Solutions, Sweden. For each tumor, maximum and mean SUV (SUV max and SUV mean ), total lesion glycolysis (TLG), and metabolic tumor volume (MTV) were determined on PET images. MTV was defined as total tumor volume with an SUV ≥ 2.5 and was calculated automatically. TLG was also calculated automatically by multiplying the MTV of the primary tumor by its SUV mean .

Histogram Analysis of ADC Values.
Automatically generated ADC maps were transferred in DICOM format and processed offline with custom-made Matlab-based application ( e Mathworks, Natick, MA) on a standard windows-operated system. e ADC maps were displayed within a graphical user interface (GUI), which enables the reader to scroll through the slices and draw a volume of interest (VOI) at the tumor's boundary (whole-lesion measure). All measurements were performed by two authors blinded to each other (AS, HJM, 15 and 2 years of radiological experience). e ROIs were modified in the GUI and saved (in Matlab-specific format) for later processing. After setting the ROIs, following parameters were calculated and written in a spreadsheet format: ROI volume (cm 3 ), mean (ADC mean ), maximum (ADC max ), minimum (ADC min ), ADC median, 10th (p10 ADC), 25th (p25 ADC), 75th (p75 ADC), 90th (p90 ADC) percentile, and mode (ADC mode). Additionally, histogram-based characteristics of the ROI-kurtosis, skewness, and entropy-were calculated.

Statistical Analysis.
Statistical analysis was performed using SPSS 23.0 (SPSS Inc, Chicago, IL). Collected data were evaluated by means of descriptive statistics. e data were not normally distributed according to Kolmogorow-Smirnow test. erefore, Spearman's correlation coefficient (p) was used to analyze associations between investigated parameters. Interreader variability was assessed with intraclass coefficients. P values < 0.05 were taken to indicate statistical significance.

Results
e investigated ADC histogram showed a good interreader variability, ranging from ICC � 0.705 for entropy to ICC � 0.959 for ADC median (Table 3). Table 4 shows results of correlation analysis between the investigated PET and ADC parameters. No statistically significant correlations were observed between SUV max or SUV mean and ADC histogram parameters.

Discussion
To the best of our knowledge, this is the first study elucidating possible correlations between ADC histogram analysis and complex 18 F-FDG-PET parameters derived from simultaneous PET/MRI in cervical cancer.
Pretherapeutic tumor staging in cervical cancer is of great importance. MRI is the best imaging modality to estimate regional tumor extent, with identification of tumor infiltration into the adjacent organs/tissues within the female pelvis [2]. Hybrid imaging, in terms of PET/CT, has been shown to be superior to other conventional imaging modalities (MRI, CT) for the identification of nodal or distant metastatic spread [21]. Consequently, the combination of both, namely, a simultaneous PET/MRI, has been described as valuable imaging modality for whole-body tumor staging of cervical cancer patients providing improved treatment planning when compared to MRI alone [22]. Furthermore, our own preliminary data show that simultaneous PET/MRI is a valuable imaging modality to reflect histopathologic parameters like cellularity and proliferation index in cervical cancer [14].
Additionally, functional MRI, as well as 18 F-FDG-PET can provide information about tumor biology in a different fashion. ADC values derived from DWI are mainly influenced by cellularity, whereas SUV values derived from FDG-PET are mainly influenced by GLUT-1 overexpression within cell membranes and enhanced activity of tumor hexokinase [4,14,23].
Presumably, parameters from PET and MRI might be associated with each other due to the fact that a more celldense tumor also might express more GLUT-1 or may have an increased enzymatic activity [9]. However, a recent metaanalysis identified only a weak inverse correlation (r � −0.30) between SUV and ADC values pooling various tumors in oncologic imaging [9]. Regarding cervical cancer, the studies, which investigated associations between ADC and SUV values, showed inconclusive results [10,[14][15][16]. Only one study found an inverse correlation between SUV max and ADC min (r � −0.532) [10], whereas most authors could not identify linear correlations between these parameters, indicating that they might reflect different tumor aspects [14][15][16].
e present study identified that several ADC histogram parameters were associated with volume-based metabolic PET parameters, namely, MTV and TLG. In good agreement with the literature, there were no correlations between ADC parameters and SUV values in the current patient sample. erefore, our results suggest that ADC histogram analysis parameters and TLG and MTV are more sensitive to reflect relationships between 18 F-FDG-PET and DWI than the widely used SUV and "conventional" ADC values. Furthermore, our study may explain negative results of the previous investigations. Moreover, in the present study, ADC values were obtained as a whole-lesion measurement, whereas in most studies [10,[14][15][16], only one slice was used for calculation and might therefore not be representative for the whole tumor. According to Kyriazi et al., whole-lesion measurement might be more beneficial than the conventional one slide approach since pixel-by-pixel ADC histograms through the entire tumor volume include different microenvironments of diffusivity, which may be masked by mean ADC analysis [24].
Furthermore, histogram-based analysis has been evaluated to have an excellent interobserver agreement [25,26]. Additionally, it could clearly discriminate between tissue affected with cancer and physiological cervical tissue [25]. Finally, it could distinguish different FIGO stages: with increasing skewness, kurtosis, and entropy in the advanced stages indicating higher tumor heterogeneity in those lesions [26].
Interestingly, ADC histogram analysis parameters correlated with some histopathological features in cervical cancer. For example, entropy was associated with p53 expression [27]. Moreover, Meng et al. identified that ADC histogram parameters can predict tumor recurrence after radiochemotherapy with an area under the curve 0.85 [28]. In another study, it was identified that skewness and several percentiles derived from ADC maps were significantly different between squamous cell and adenocarcinomas of the uterine cervix and, therefore, ADC histogram analysis might aid in discrimination of the entities [29]. In fact, as reported previously, skewness was significantly higher for squamous cell carcinomas than adenocarcinomas and was higher in poorly differentiated tumors [29].
Regarding 18 F-FDG-PET, pretreatment SUV max and MTV have been reported to be associated with tumor prognosis [30,31]. So MTV had a hazard ratio of 3.15 for disease-free survival [31], and SUV max of the primary tumor was the only identified prognostic factor in a multivariate analysis [30]. Furthermore, TLG was also associated with the overall survival in locally advanced cervical cancer [32]. However, it might be of limited use for primary diagnosis in early stage carcinomas since 18 F-FDG-PET only has little value in the routine pretreatment assessment in patients with early FIGO stages [33]. However, there are promising histopathological methods to better understand underlying microstructure changes, which can be displayed with PET imaging [34].

Contrast Media & Molecular Imaging
Overall, our report indicates that for further analyses about associations between DWI and PET and as well between PET, DWI, and histopathology in several tumors, ADC histogram analysis and volume-based metabolic PET parameters like TLG/MTV should be obtained.
ere are several limitations of the present study to address. Firstly, it is a retrospective study with possible known bias. However, MRI and 18 F-FDG-PET were measured by two different readers, blinded to each other. Secondly, the patient sample is relatively small. irdly, only squamous cell carcinomas were evaluated.
In conclusion, the present study shows that ADC histogram analysis and volume-based metabolic 18 F-FDG-PET parameters are related to each other and might, therefore, reflect similar tumor behavior of cervical cancer. e next step would be to assess the value of these simultaneous PET/MRI parameters for predicting treatment response and survival in cervical cancer patients.

Data Availability
e anonymous patient data used to support the findings of this study are available from the corresponding author upon request.

Conflicts of Interest
e authors declare no conflicts of interest.

Authors' Contributions
Hans-Jonas Meyer, Alexey Surov, and Sandra Purz wrote the manuscript. Hans-Jonas Meyer and Alexey Surov performed histogram analysis. Sandra Purz and Osama Sabri performed PET analysis. Hans-Jonas Meyer performed the statistical analysis. All authors contributed equally to this work.