Association Between Diffusion Weighted-Imaging (DWI) and Simultaneous 18F-FDG-PET/MRI Parameters with a Comparison of their Diagnostical Role in Head and Neck Squamous Cell Carcinoma (HNSCC)

Background: Hybrid PET/MRI is an emerging imaging technology proved to be useful for better understanding of the tumor metabolism and cellularity, it also plays a very important in staging, assessment and post-therapy follow up. PET/MRI can be used to better understand how tumors act, especially prior to therapy. Our aim in this study is to assess the association of 18 F-Fluorodeoxyglucose positron-emission-tomography (18F-FDG/PET) and DWI imaging parameters and multi-clinical factors correlations and comparing their diagnostical performance to predict tumor aggressiveness in HNSCC. Results: No signicant correlations were found between DWI and any of 18 F-FDG parameters SUVmax, TLG and MTV, (r = -0.184, P=0.125, r = -0.182, P=0.248, and r = -0.037, P=0.756), respectively. As SUVmax and TLG of the primary tumor increase, the tumor aggressiveness to involve more lymph nodes increase, (r = 0.321, P=0.006 and r = 0.332, P=0.005), respectively. Comparison between patients with positive (N+) and negative (N-) lymph node groups show that SUVmax and ADC can predict lymph nodes metastasis, (P=0.004 and P=0.012), respectively. SUVmax best cut-off value of (6.8±0.8), had higher accuracy than ADC, best cut-off value of (0.981±0.97*10 -3 mm 2 /s), (sensitivity: 83.6%, 70.0% and specicity: 80.0%, 78.7%), respectively. Additionally, TLG and MTV were positively correlated with T-stages (P=0.024 and P=0.001), respectively. ADC was inversely correlated with tumor grades (P=0.030). Conclusions: Our results revealed a non-signicant correlation between the FDG-PET and DWI-MR parameters. The FDG-PET-based glucose metabolic and DWI-MR derived cellularity biological aspects of HNSCC. superior to in predicting lymph metastasis. Table Final conrmation of malignancy was done after PET/MRI the and metastatic lymph nodes combined with biopsy. Intravenous 18 F-FDG a bodyweight adapted (4 MBq/kg, range 163–403 MBq) intravenously injected; after the FDG tracer injection, acquisition started within 142 minutes (average 225 minutes). were obtained in the supine position using Head and Neck coils. MRI sequences T2-weighted TSE turbo inversion recovery magnitude (TIRM) 3300/37/220 ms, FOV: 240 mm, slice 3 mm, 224 × 320) coronal plan, T1-weighted turbo spin-echo (TSE) (TR/TE 800/12 ms, FOV: 200 mm, slice 4 mm, 224 × 320) and T1-weighted TSE Dixon fat (FS) (TR/TE 6500/85 ms, FOV: 200 mm, slice thickness: 4 mm, 256 × 320) transversal and were acquired without an intravenous contrast agent. For the PET data collection, a magnetic resonance-based attenuation correction (MRAC) sequence was used for PET attenuation correction, and the wide range bed position PET Emission scan was acquired for 900 seconds with a xed FOV range (20 cm) and matrix (172 × 172) without bed movement as well. An iterative ordered subset expectation maximization (3D OP-OSEM) PET image reconstruction algorithm was used with 3 iterations and 8 subsets, and 4 mm Gaussian ltering settings. The PET data were corrected for scattering, random coincidences and attenuation using the MR data. The DWI was obtained by using an axial echo-planar imaging (EPI) sequence with b-values of 0 and 800 s/mm 2 (FoV 315 mm, repetition time TR/TE: 9900/75 ms, 5 mm slice thickness and voxel size 2.3

[8], [9] The higher cellular tumor resulted in more restriction to water molecule motion which, as a result, gives lower ADC values and vice versa. [10] This means that the water molecule's motion is re ecting the signal loss on DWI due to different water permeability through the structures. [11] Previous studies have proved the inversely proportional correlation between ADC and tumor cellularity. [12], [13] ADC also was found to be effective in primary tumor assessment, differentiating between benign and malignant neoplasms, staging and monitoring post-treatment followup. [14], [15] Moreover, ADC was found to be useful for predicting treatment response in HNSCC. [16] The FDG uptake values measured from PET imaging has an important role in head and neck imaging due to its ability to measure the glucose metabolism in the tumors, [17]- [19] which may also re ect the tumor's aggressiveness and the risk of the metastasis to spread to the adjacent structures. [20], [21] SUV is the most common parameter used to estimate glucose metabolism, and it has shown promising results in predicting the presence of lymph nodes metastatic during the primary assessment as well as a predictor of survival and recurrence. [22] Recently; new metabolic parameters, TLG and MTV have emerged as new parameters that can measure the glucose metabolism activity of tumors and have been founded to be more effective than SUV because tumor contour is considered when using MTV and TLG. [23] Since SUVmax doesn't re ect the metabolic activity of the entire lesion but it measures the highest glucose metabolism in the target ROI. [24] While MTV represents the volume of the 18 F-FDG activity in the lesion and TLG represents the sum of the SUV within the lesion. Furthermore, the glucose metabolic activity is positively correlated to the tumor cellularity. [25], [26] Previous studies suggest that tumor cellularity and metabolism might be correlated. However, previous results were discordant; Varoquax et al. found, in their study of SCC, that there was no signi cant correlation between the tumor cellularity represented by ADC and the tumor metabolic activity represented by SUV. [27] Fruehwald et al. reported that there was no correlation between SUV and ADC either in the DWIBS or EPI.
[28] Similar ndings have been reported by other authors. [29]- [31] In contrast, Nunez et al. reported in their study of HNSCC that the metabolic activity was strongly correlated to the tumor cellularity; there was an inverse signi cant correlation between the ADC and SUV. [32] Nakajo et al. found that the tumor metabolic activity (SUV) was correlated inversely with the tumor cellularity represented by ADC. [21] It's not clear yet why some authors have found strong correlations while others have not, and whether there is a correlation between tumor cellularity and metabolic activity.
18F-FDG imaging parameters and DWI'ADC are a commonly used parameter in PET/MRI. These imaging parameters show a promising results to measure activity level of tissue metabolism, cellularity and proliferation. [9], [33] The use of these imaging parameters were also expanded to study the differences between benign and malignant in the microstructure level, prediction of treatment response, survival analysis and their correlation with the clinical and pathological information of the tumors. [14], [22], [34] Therefore, our study was aimed to investigate the correlation between FDG parameters and ADC values, which has focused, in-depth, on nding out if there is a correlation between tumor metabolic activity and cellularity represented by ADC and SUVmax, TLG and MTV, as well as assessing the ability of these imaging parameters to determine tumor aggressiveness by predicting lymph nodes involvement.

Materials And Methods:
Patients and demographics: A retrospective study was approved by the Clinical Center, Regional and Local Research Ethics Committee (CCRLREC), Doctoral School of Health Sciences, University of Pecs, and Somogy Megyei Kaposi Mor Educational Hospital, Pecs, Hungary. Approval number (IG/00686-000/2020). Requirement of the informed consent was waived and con rmed by the (CCRLREC) due to the retrospective nature, and all methods were carried out in accordance with the relevant guidelines and regulations (Declaration of Helsinki). From May 2016 to June 2019, 109 patients with proven HNC underwent 18 F-FDG PET/MRI for staging and restaging, assessment of the disease and post-therapy follow up. The inclusion and exclusion criteria were (1) proved non-treated primary HNC, (2) patients underwent PET/CT and PET/MRI including DWI sequence (3) single tracer injection session. Exclusion criteria (1) patients who had nonmeasurable ADC, or FDG parameters (2) patients with motion artifact or bad image quality. Finally, a total of 71 patients were included in our study. Table (1). Final con rmation of malignancy was done after PET/MRI examination the primary tumor and metastatic lymph nodes combined with biopsy. Image analysis: In each patient, the SUVmax, TLG, MTV were measured from the PET imaging; Siemens (Syngo Via 10VB) was used, which provided an automatized delineated SUV-based volumetric analysis. The metabolic volumetric contours were segmented by using the Syngo Via (VOI) Sphere tool. The single voxel activity concentration of a particular tumor with the highest SUV was represented by SUVmax while SULpeak represented the hottest point in the tumor foci, where the lean body mass normalized as the average SUV was measured at 1 cm 3 in a spherical ROI. A xed 2.5 threshold of SUV was used for tumor SUVmax for both MTV and TLG proposed by Pak et al. [35] The volume above the given VOI was represented the MTV while the TLG represented the VOI of the average SUVmean or SULmean multiplied by the MTV. The ADC map was automatically generated and analyzed on the implemented eRAD software. DWI images were analyzed by drawing round or oval region of interest (ROI) manually on the ADC map covering the largest tumor diameter, [18] on single DWI slice [28] within the center of the lesion in the most homogenous part which were the lowest ADC or the highest SUV reported after excluding or/and avoiding the necrotic and cystic areas. We did not use whole tumor volumes ADC measurements approach although it has been found to be more reproducible than those obtained from single slice or small ROI's measurements.
However, there was no signi cant difference between the tumor ADCs obtained using whole-volume measurements and the single-slice approach.
[36] Thus, we have chosen the single-slice method because it's easier, faster and as a result more preferred in clinical practice than the whole volume ROIs protocol which is time consuming and more complicated. Average ADC values calculated by the software automatically was referred to as ADCmean by summing all voxels ADC values on the drawn ROI for the chosen slice. We assessed only ADCmean values, which as previously proposed as a more reliable indicator of tumor cellularity since the entire lesion is taken into account. [37] ADCmin, on the other hand, was suggested to re ect the most proliferative portion of a tumor or highest tumor cell density, due to the effects of lesion heterogeneity or artifacts the use of ADCmin is likely to result in more errors.
[38] In addition, ADCmean minimizes the effect of tumor heterogeneity and its higher reliability to distinguish different entities in the same image. [39] We used the average ADC of the overall area included in the ROI which is calculated automatically by the software, where "Avg" represents the average ADC values for all voxels within the ROI and "Dev" Represents the standard deviation. Figures (1).

Statistical analysis:
Statistical analysis was performed by using SPSS 25 (IBM SPSS Statistics, Armonk, New York, USA). The data collected were evaluated using descriptive statistics (mean ± standard deviation), for variables with normal distribution and median and interquartile range for variables with non-normal distribution. The Spearman rank correlation (r) was used to estimate the association between 18F-FDG parameters and DWI values as well as tumor size, T stages, N stages and tumor grades. ANOVA or Kruskal-Wallis test were performed with primary tumor localization. Independent sample t or Mann-Whitney test were applied to compare imaging parameetrs values with Sex, M stages. Variables for which P < 0.1 in univariate analysis were subjected to multiple linear regression analysis to determine those that were independently associated with the imaging parameters by integrating statistically differences in the univariate analysis into the multivariate linear regression model, we used transforming function to convert variables with non-normal distribution into normal distribution. Mann-Whitney test and independent sample T-test were applied on the imaging parameters after the patients were grouped based on lymph nodes involvement into positive (N+) and negative lymph nodes (N-). Receiver operating characteristics (ROC) was recruited   [40] localization and the degree of differentiation (grades). The results show that SUVmax was correlated positively with tumor size and N stages, (P = 0.001 and P = 0.006), respectively. TLG was positively correlated with tumor size, T stages and N stages (P = 0.000, P = 0.024 and P = 0.005), respectively. MTV was positively correlated with tumor size and T stages, (P = 0.000 and P = 0.001), respectively. ADC, in the other side, was found to be inversely correlated with the degree of differentiation (P = 0.030) and a tendency to correlate with N stages, (P = 0.089). No other signi cant correlations observed, (P > 0.05) for all parameters. Table (3).  .

Signi cant result was highlighted in Bold
Multiple regression analysis was recruited for factors that shown correlation (P < 0.1) in univariate analysis to investigate the factors that in uence the change in SUVmax, TLG, MTV and ADC. The results show that tumor size and N stages were independent factors in uencing SUVmax, (P = 0.020 and P = 0.024), respectively. Tumor size and N stages were independent factors in uencing TLG, (P = 0.000 and P = 0.044), respectively. T stages and tumor size were independent factors in uencing MTV (P = 0.004 and P = 0.000), respectively. Tumor grade was found to be independent factor in uencing ADC (P = 0.032).  When excluding the effect of the tumor size from the regression model, we found that N stages were independent factor in uencing SUVmax (P = 0.011), but not T stages (P = 0.838). Both T stages and N stages were independent factors in uencing TLG, (P = 0.018 and P = 0.034), and T stages were found to be independent factor in uencing MTV, (P = 0.001).
To investigate the ability of FDG and ADC parameters to predict tumor aggressiveness, we classi ed the patients based on lymph nodes involvement into Negative and Positive groups (N-and N+) and compared with these parameters. PET/MRI was the reference to de ne the two groups. Our results show that SUVmax, TLG and ADC revealed a statistically signi cant differences (P = 0.004, P = 0.033 and P = 0.012), respectively. MTV did not (P > 0.05). Figure 3 (A, B and C and D).
The ROC curve was used to analyze the diagnostic e cacy of ADC and SUVmax (due to widely use in daily practice) in predicting lymph node metastasis in HNC. For ADC; AUC was 73.1%, 95% con dence interval was ranged between 0.550 and 0.912, best cut off value was (0.981 ± 0.97) to predict lymph node metastasis with sensitivity of 70.0% and speci city of 78.7%, Fig. 4 (A). SUVmax best cut off value to predict lymph node metastasis was (6.8 ± 0.8), AUC was 80.8%, 95% con dence interval was ranged between 0.633 and 0.984. Sensitivity and speci city were 83.6% and 80.0%, respectively. Figure 4 (B).

Discussion:
The present study demonstrated that PET/MR provides valuable imaging data for HNC patients. Various pathological factors were associated with PET/MR results and may serve a role in the evaluation of the prognosis of patients with HNC. As for emerging technology, PET/MRI offers different imaging data to study tumor microstructure environment, we started our study by correlating these imaging to each other. Previous data demonstrated an inverse correlation between ADC value, derived from DWI, with cellularity.
[8]- [10] FDG imaging parameters, on the other hand, were found to be positively correlated with cellularity. [25], [26], [41] Although glucose metabolism and cellularity of tissue are two different biological biomarkers of a tumor, an inverse correlation between 18F-FDG and DWI parameters has been suggested. [42] this hypothesis was proposed because both 18F-FDG and ADC were correlated with tumor cellularity. [37] Our results showed that FDG uptake parameters (SUVmax, TLG, and MTV) were not signi cantly correlated with the ADCmean value. Similar results were observed; Min et al., in their study of HNSCC, reported that there was no signi cant correlation between ADCmean with SUVmax and SUVmean, also no signi cant correlation was found between ADCmean and both MTV and TLG. Our explanation for the lack of correlation is the fact that both imaging parameters explain different tissue microstructures characteristics, DWI assess the water molecule motion in the tissue and affected by the cellularity, proliferation rate and cell counts which in clinical use affected by ROI size placement and interobserver variability.
[36] While metabolic activity was independent of tumor size and shape because tumor is segmented by adaptive thresholding. [37] The present study correlated FDG and DWI imaging parameters with cliniopathological characteristics to explore their effect on the imaging parameters values. Our results reveal that FDG metabolic parameters have reported different correlations; it has shown that primary tumor SUVmax and TLG were signi cantly correlated with tumor size and N stages; the larger tumor size means more cancer cells, thus, more active overall hyperplasia, in other words, greater glucose metabolic activity to tolerate the biological activity, differentiation and proliferation of the cancer cell. [45] Metastatic lymph node, in the other hand, is one of the most important in uencing factors in the prediction of cancer surgery. [1] It's well known that higher degree of malignancy means more in ltration and thus the possibility of lymph nodes metastasis is high.
[46] According to El-naaj et al. when there is no noticeable lymph node metastasis in the clinical and imaging examinations, the incidence of occult metastasis was high (20-34)%. [47] Thus, it's important to predict the possibility of lymph node metastasis occurrence. This study was found that the AUC was 0.808, which means that SUVmax is useful to predict lymph node metastasis. According to Zheng  ADC and nodal involvement, (P = 0.003), [50] this mean that patients without lymph nodes involvement showed higher ADC value than those patients who have con rmed lymph nodes enlargement. In the other hand, Nakajo et al. have reported in their study of primary HNSCC similar results, there was no signi cant difference in the ADC between N-positive and N-negative groups (p = 0.74), [21] similar results were also reported by other authors. [51], [52] The explanation of their result was attributed that those patients with poorly or undifferentiated malignancy are usually reporting metastatic lymph nodes. [50] None of the previous studies have compared the e cacy of PET/MRI system different imaging biomarkers in HNC tumor aggressiveness prediction. Thus, to our knowledge, this is the rst study to compare PET/MRI system derived imaging parameters in lymph nodes involvement in HNSCC. Our results show that SUVmax and ADC were found to have the ability to differentiate between the two lymph nodes groups (N + and N-) based on the primary tumor measurements, which as a result might help to predict tumor development and prognosis. Our study shows that SUVmax had higher diagnostic performance, higher sensitivity and better speci city than ADC, as well as the presence of signi cant correlation with N stages which has not been found in ADC. Nevertheless, ADC can predict tumor aggressiveness and lymph nodes involvement prediction but with limited e cacy. The importance of successful prediction of tumor aggressiveness and lymph nodes involvement might help in practice to increase the aggressiveness of the therapy.
Based on our study results and ndings, there were several correlations between PET/MRI imaging parameters and clinical tumor characteristics, we suggest that glucose metabolism assessed by 18F-FDG and cellularity assessed by ADC have different roles in cancer evaluation, so we recommend PET/MRI as a combined examination rather than PET or MRI alone.
As for this study's limitations, First, the heterogeneity of the tumor localization. Second, our study focused on the search of correlation between 18F-FDG, ADC and histopathological features only in HNSCC. Third, associations with other functional tumor parameters, such as apoptosis factors and were not analyzed.
Fourth, design of the study was retrospective.

Conclusion:
Our results revealed no linear correlation between the FDG PET and DWI-MR parameters. The FDG PET- Declarations that greatly assisted the research, although they may not agree with all of the interpretations of this paper.
The Authors' Contributions: OF designed the study, OF and PT collected and processed the data. DS and AK segmented FDG measurements and generated the gures. OF conducted data collection and processing, statistical analysis and wrote the paper. TZ review the draft. AK, ZS, and IR discussed the results and contributed to the nal form of the article.   Receiver operating characteristic (ROC) curve analysis of lymph nodes prediction according to ADC and SUVmax of primary tumor. (A) ADC (ROC) curve with AUC (73.1%), 95% con dence interval was ranged between 0.550 and 0.912, best cut off value was (0.981±0.97*10-3mm2/s) to diagnose lymph node metastasis with sensitivity of 70.0% and speci city of 78.7%. (B) SUVmax (ROC) curve with AUC was 80.8%, 95% con dence interval was ranged between 0.633 and 0.984, best cut off value to diagnose lymph node metastasis was (6.8±0.8) with sensitivity (83.6%) and speci city 80.0%.