Multiparametric MRI Evaluation of Oropharyngeal Squamous Cell Carcinoma. A Mono-Institutional Study

The aim of this paper is to define the pre-treatment radiological characteristics of oropharyngeal squamous cell carcinoma (OPSCC) using morphological and non-morphological magnetic resonance imaging (MRI), based on HPV status, in a single-institution cohort. In total, 100 patients affected by OPSCC were prospectively enrolled in the present study. All patients underwent 1.5T MR with standard sequences, including diffusion-weighted imaging with and intravoxel incoherent motion (IVIM-DWI) technique and a dynamic contrast-enhanced (DCE) MRI. For all patients, human papillomavirus (HPV) status was available. No statistically significant differences in the volume of primary tumors (PTs) and lymph nodes (LNs) were observed based on HPV status. When comparing the two patient groups, no significant differences were found for the PT radiologic characteristics (presence of well-defined borders, exophytic growth, ulceration, and necrosis) and LN morphology (solid/cystic/necrotic). Tumor subsite, smoking status, and alcohol intake significantly differed based on HPV status, as well as ADC and Dt values of both PTs and LNs. We detected no significant difference in DCE-MRI parameters by HPV status. Based on a multivariate logistic regression model, the combination of clinical factors, such as tumor subsite and alcohol habits, with the perfusion-free diffusion coefficient Dt of LNs, may help to accurately discriminate OPSCC by HPV status.


Introduction
Squamous cell carcinoma (SCC) is a common malignant tumor affecting the skin. A peculiar feature of this tumor is the abnormal and quick growth of keratinocytes in the epidermis, often secondary to ultraviolet or sunlight exposure [1]. When this tumor involves the head and neck area, it may be particularly aggressive and can account for up to 90% of all oral malignancies [2]. In recent decades, there has been a dramatically increased incidence in oropharyngeal squamous cell carcinoma (OPSCC) associated with human papillomavirus (HPV) infection [3].
At present, HPV-related OPSCC has been found to be a separate subtype of head and neck squamous cell carcinoma (HNSCC), characterized by peculiar biologic features, clinical course, and response to treatment [4]. HPV-OPSCC has been shown to have a better prognosis [5]. A meta-analysis of retrospective studies by Ragin et al. [6] indicated that patients with HPV-positive OPSCC exhibited markedly reduced risk of mortality compared to patients affected by HPV-negative OPSCC. Fine needle aspiration cytology (FNAC) with ultrasonography, computed tomography (CT), and MRI are the techniques generally used for the local staging of OPSCC, while positron emission tomography (PET)-CT is useful in locally advanced stages and for carcinomas of unknown primary origin [7][8][9][10][11]. Several studies [11][12][13][14] have reported differences in imaging features of OPSCC by HPV status. HPV-related lymph nodes (LNs) often tend to present with large cystic masses, with a clinically and radiologically smaller or even occult primary tumor (PT) [13]. In recent years, there has been an increasing focus on evaluating OPSCC with a multiparametric MRI approach using non-morphological imaging, i.e., diffusion-weighted imaging (DWI) and perfusion-weighed dynamic contrast-enhanced MRI [15][16][17]. DWI allows for the gathering of information about cellularity, and the apparent diffusion coefficient (ADC) has been found to correlate with cell density and proliferation [16]. DCE provides information about tissue vascularity and is also related to various histopathological features, such as vessel count, total vessel area, and microvessel density [17,18]. The possibility of predicting HPV status by imaging biomarkers can be used as a non-invasive and cost-effective tool in patients who have undergone imaging for staging.
In this study, we analyze the pre-treatment MR characteristics of OPSCC in a singleinstitution patient cohort. In particular, we focus on morphological and non-morphological MRI features of both primary tumors and neck nodes, including diffusion-weighted and perfusion-weighted sequences, after separating patients by HPV status.

Patients
The study protocol was authorized by the local Institutional Review Board (approval number RS716/15), and specific, informed consent was obtained from each patient. The criteria for inclusion were: (i) age older than 18 years; (ii) OPSCC confirmed by biopsy; and (iii) availability of HPV status. Exclusion criteria included: (i) any contraindication to MR examination; (ii) prior surgery as excisional biopsy of the tumor or neck nodes; (iii) prior chemotherapy (CHT), including induction CHT or radiotherapy to the primary disease and the neck; (iv) unknown HPV status; and (v) distant metastases at presentation (M1). The disease was staged following the 8th edition of the AJCC (American Joint Committee on Cancer) Cancer Staging Manual for HNSCC [19].

HPV Detection
Two techniques (p16 immunohistochemistry and PCR-based detection) were used to identify HPV-positive patients, defined as those with lesions positive for both p16 and HPV-DNA [20]. Depending on tissue extent, a variable number of 1−3 × 5 µm sections was cut for each formalin-fixed paraffin-embedded (FFPE) cancer tissue. HPV-DNA detection and genotyping were achieved by PCR-based INNO-LiPA HPV Genotyping Extra II kit (Fujirebio, Malvern, PA, USA) and a TENDIGOTM instrument (Fujirebio, Malvern, PA, USA), after purifying DNA with a DNeasy Blood and Tissue Kit (Qiagen, Hilden, Germany). The expression of p16 protein was determined by a CINtec ® Histology Kit (Roche Diagnostics, Milan, Italy).

MR Imaging Protocol
MRI examinations were performed on a 1.5-T system (Optima MR 450w, GE Healthcare, Milwaukee, WI, USA) with 16-channel radiofrequency coils, including a head coil, a surface neck coil, and a spine coil. The MR examination included coronal fast spin-eco (FSE) T2-weighted (T2-w) images (slice thickness, 4 mm), followed by axial FSE T2-w images (slice thickness, 3 mm), and pre-contrast axial T1-w images (slice thickness, 3 mm), acquired from the level of the skull base to the thoracic inlet. DWI were obtained via single-shot spin-echo and echo-planar imaging (slice thickness, 4 mm), with multiple b values (b = 0, 25, 50, 75, 100, 150, 300, 500, and 800 s/mm 2 ). DCE-MRI was obtained through a 3D fast-spoiled gradient echo sequence (slice thickness of 6 mm, and spacing between slices 1.5 mm). In total, 30 dynamic volumes were acquired after having optimized temporal resolution (5 s) and total scanning time (5 min and 15 s). After three dynamic volumes, a contrast agent (0.1 mmol/kg bodyweight of gadopentetate dimeglumine) was injected intravenously, with a rate of 3 mL/s. According to the standard acquisition protocol, the protocol also included post-contrast T1-w images with liver acquisition with volume acceleration sequences (LAVA; slice thickness 1 mm, 214 slices) on axial and coronal planes.

Image Analysis
PTs and LNs were identified and manually contoured by two expert HN radiologists (F.P. and E.G.) in consensus. Anatomical T2-w images and/or post-contrast T1-w images were used for PT and LN segmentation and volume size quantification. Image visualizing and contouring were performed using 3D Slicer Software (Version 4.1.1). Based on their appearance on T2-w images and/or post-contrast T1-w images, PTs were described as having well-defined borders, as well as being exophytic, ulcerated, and necrotic, while LNs were classified as solid, cystic, or necrotic. Cystic change was defined as a homogeneous low signal intensity on T1-w images and high signal intensity on T2-w images relative to normal muscle, with the enhancement of the capsule after contrast medium with welldefined margins, homogeneous fluid content with no internal complexity, irregularity, or solid component by criteria of Goldenberg et al. [14], or with a component greater than 20% of the total volume by the criteria of Morani et al. [13]. In the case of irregular margins, it was considered a necrotic focus [13]. Radiologic extracapsular spread (rECS) of LNs has been described as irregular margins with invasion or distortion of surrounding soft tissues [13].
The diffusion parameters were derived from homemade scripts created in the MAT-LAB programming language (Release 2018b, The MathWorks Inc., Natick, MA, USA). To perform the data fitting, the median value of the signal within the entire lesion was calculated at each b value. Data points acquired at b values of 300, 500, and 800 s/mm 2 were used to estimate the tissue diffusion coefficient D t (in mm 2 /s) based on a mono-exponential function; ADC was similarly calculated using b values of 0, 500, and 800 s/mm 2 [21]. The perfusion-related coefficients, f (fractional volume of capillary blood) and D * (perfusionrelated diffusion coefficient), were not included in the analyses, considering that, in a previous investigation [15], no association was found between these variables and the HPV status.
Perfusion maps were derived from commercial software (GenIQ General, GE Advanced Workstation, Palo Alto, CA, USA). A pharmacokinetic two-compartment model was applied to DCE-MRI images to extract: K trans , the transfer constant between plasma and the extravascular extracellular space (EES); K ep , the transfer constant between EES and plasma; and v e , the fractional volume of EES [22]. The semi-quantitative parameter, IAUGC, which represents the initial area under the gadolinium concentration curve (calculated from the bolus arrival to the first 90 s), was also estimated. Particular attention was paid to remove arteries/veins and bones from PT/LN contours. The median of each parameter within the total lesion volume was used to perform the statistical tests.

Statistical Analyses
Median values and interquartile ranges were used to describe continuous variables, while frequencies and percentage values were used for categorical variables. Differences between continuous variables were assessed with the Mann-Whitney test or Student's t-test, as appropriate, while the relationships between categorical variables were evaluated using the chi-square or Fisher's Exact tests, as appropriate. The Shapiro-Wilk test was applied to evaluate the normality distribution of the data.
The chi 2 test was used to evaluate potential differences in smoking status and alcohol intake by HPV status. Specifically, patients were defined as non-smokers if they smoked <6 packs/year, moderate smokers if they smoked 6 ≤ packs/year < 25, and strong smokers if they smoked ≥25 packs/year; patients were defined as a moderate alcoholic if they drank <1 L of wine/day and heavy alcoholic if they drank ≥1 L of wine/day.
Additional tests were conducted considering three patient subgroups according to smoking status, alcohol intake, and HPV status: specifically, HPV-negative patients, HPVpositive non-smoker (<5 pack/year), non-alcoholic patients and HPV-positive smokers (≥6 pack/year), and moderate or heavy alcoholic patients. The Kruskal-Wallis test was applied to assess the differences between the groups of patients.
A univariate logistic regression model was applied to identify variables that might play a role in the risk of HPV status in OPSCC. A multivariate logistic regression model was used with predictive variables that were significant in the univariate analyses following the forward selection method. The confidence interval for the area under the receiver operating characteristic curve (AUC) of each model was calculated with the bias-corrected and accelerated percentile bootstrap method, using 1000 replicates. The model classification ability was also evaluated in terms of accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV).
The statistical calculations were performed using SPSS statistical software version 21 (SPSS Inc., Chicago, IL, USA).

Results
In total, 100 patients affected by OPSCC, 72 HPV-positive and 28 HPV-negative, were prospectively included in our investigation from January 2016 to September 2019. Table 1 reports the selected patient and tumor characteristics. Patients with HPV-related OPSCC tended to be younger and were more likely to have lower N-stage disease than patients with non-HPV-related OPSCC. The proportion of HPV-positive OPSCC was significantly affected by the tumor subsite (p = 0.025).
Data on smoking status were missing in one HPV-negative patient, while data on alcohol intake were missing for two HPV-negative patients and one HPV-positive patient. Regarding smoking status, 31 HPV-positive and 4 HPV-negative patients were non-smokers; 13 HPV-positive and 4 HPV-negative patients were moderate smokers; and 22 HPV-positive and 25 HPV-negative patients were strong smokers (p < 0.001). Regarding alcohol intake, 45 HPV-positive and 9 HPV-negative patients were non-drinkers; 19 HPVpositive and 6 HPV-negative patients were moderate alcoholics; and 4 HPV-positive and 14 HPV-negative patients were heavily alcoholic (p < 0.001).
No statistically significant differences in PT volumes (median value of 11.6 cm 3 for HPV-positive group and 17.8 cm 3 for HPV-negative group; p value = 0.539) and LN volumes (median value of 5 cm 3 for HPV-positive group and 7 cm 3 for HPV-negative group; p value = 0.317) were observed.

Multiparametric MRI Evaluation
When comparing the two patient groups, no significant differences emerged for PT characteristics (presence of well-defined borders, exophytic growth, ulceration, and necrosis) or the morphology of LNs (Table 2). The distribution of rECS was similar in HPV-positive and HPV-negative patients, with 38 cases (60%) of ECS and 14 (54%), respectively (chi-square test, p = 0.107).
Conversely, ADC and D t significantly differed by HPV status in PTs (p = 0.009 and p = 0.001, respectively) and LNs (p = 0.007 and p = 0.029, respectively), as shown in Table 3.
No statistically significant differences were reported in DCE perfusion parameters (K trans , K ep , v e , and IAUGC) between the two patient groups for both PTs and LNs (Table 4). Table 3. ADC and D t coefficients in primary tumors (PTs) and metastatic lymph nodes (LNs).

Interactions between Imaging-Based Parameters and Clinical Factors
Some interactions between imaging-based parameters, particularly ADC and D t , and the most relevant clinical factors associated with HPV status were evaluated. We investigated whether the tumor subsite (tonsil or base of the tongue) could significantly impact ADC and D t (Supplementary Materials, Table S1). We found that the tumor subsite significantly affected ADC and D t parameters of PTs (p = 0.001 and p < 0.001, respectively) but not ADC and D t parameters of LNs (p = 0.274 and p = 0.243, respectively). Based on this, the differences in ADC and D t of PTs by HPV status were further investigated in tonsil and base of the tongue groups, separately (Supplementary Table S2). Even though both ADC and D t values of HPV-positive PTs were still lower than those of HPV-negative PTs, only in the base of the tongue was the difference in D t on the significance threshold (p = 0.051).
Analogously, a comparison between diffusion parameters of HPV-negative, HPVpositive non-smoker, non-alcoholic patients, and HPV-negative versus HPV-positive smoker and alcoholic patients, for both PTs and LNs, was performed (Supplementary  Table S3).
Lastly, we assessed whether both ADC and D t could be considered independent from tumor size and rECS (Supplementary Tables S4 and S5). No (or a very weak) association was found between ADC/D t and volume size, both for PTs and LNs, and ADC/Dt values did not significantly differ between LN with or without rECS. This suggested that tumor size and LN tumor invasion did not significantly impact the ADC and D t parameters, independently of the HPV status of the tumor.

Logistic Regression Models
Univariate and multivariate logistic regression analyses for HPV positivity are reported in Table 5, including the two larger subgroups of the tumor subsite (base of the tongue, n = 42 and tonsil, n = 53). Due to the high collinearity between the ADC and D t parameters, only D t was included in the multivariate model.
Two illustrative cases, respectively, of an HPV-positive OPSCC of the tonsil in a nonsmoker and non-alcoholic man, and of an HPV-negative OPSCC of the base of the tongue in a moderate smoker and alcoholic man, are presented in Figures 2 and 3.

Discussion
Recently, several investigations have focused on the dissimilar characteristics of OPSCC in relation to HPV status, using multiple techniques, i.e., CT, MRI, and PET-CT, to obtain imaging biomarkers related to lesion volume size, margins, tumor microenvironment, cellular density, vascular perfusion, and metabolism [22][23][24][25][26][27]. However, to our knowledge, this is the first study proposing a comprehensive evaluation of both anatomical and functional characteristics, based on multiparametric MRI, together with patient-related factors.
Consistent with previous studies, we reported that patients with HPV-related OPSCC had different smoking and drinking habits than HPV-negative patients [28].
Regarding morphological features in our population, we did not observe significant

Discussion
Recently, several investigations have focused on the dissimilar characteristics of OP-SCC in relation to HPV status, using multiple techniques, i.e., CT, MRI, and PET-CT, to obtain imaging biomarkers related to lesion volume size, margins, tumor microenvironment, cellular density, vascular perfusion, and metabolism [22][23][24][25][26][27]. However, to our knowledge, this is the first study proposing a comprehensive evaluation of both anatomical and functional characteristics, based on multiparametric MRI, together with patient-related factors.
Consistent with previous studies, we reported that patients with HPV-related OPSCC had different smoking and drinking habits than HPV-negative patients [28].
Regarding morphological features in our population, we did not observe significant differences in terms of PT characteristics between the two groups; although, in HPVpositive patients, lesions with well-defined and exophytic margins were slightly more prevalent, as reported by Chan et al. [24] using both CT and MRI, and by Cantrell et al. [11] based on CT imaging.
Regarding the morphological features of OPSCC neck nodes, differences have been reported by HPV status, with the HPV-positive group showing a cystic appearance that was slightly predominant [11,13,[29][30][31]. On the contrary, in our series, we found a majority of solid neck metastases for both patient groups. This aspect is important to predict the risk of treatment failure, as suggested by Rath et al. [12], who found that, in HPV-positive patients, solid neck metastases represented an independent element for treatment local failure after chemoradiation therapy together with T-stage and ECS. In another study by Barakat et al. [32], it was noticed that cystic nodes metastasis showed a shrinkage to solid ones after induction chemotherapy.
Regarding rECS, in our series, we did not find a statistically significant difference between HPV-positive and negative patients. In terms of the significance of the ECS, in patients with HPV-positive OPSCC, there are contradictory opinions: some authors [33,34] maintain that ECSs are not correlated with regional recurrence or distant metastases, while, more recently, other authors [23,35] have argued that ECSs may be associated with increased rates of locoregional recurrence and distant metastasis, with worse overall survival, and with disease-free survival. In a recent review of Park et al. [35], who evaluated the diagnostic performance of CT and MRI, sensitivity and specificity were 73% and 83% for CT and 60% and 96% for MRI, respectively; however, a lower specificity was found in patients affected by HPV-positive OPSCC (74% versus 87%), with no significant difference in sensitivity.
The recent literature demonstrates a growing interest in using multimodality techniques to obtain a quantification of functional features of tumors that are potentially associated with tissue architecture, microvascularity, or metabolic activity [29,36], intending to offer personalized treatment and management strategies [37]. Moreover, identifying HPV status from imaging biomarkers may be useful in the case of lesions hardly accessible for biopsy, as well as if a patient declines invasive procedures such as transoral robotic surgery.
The potential of DWI to discriminate OPSCC by HPV status has been already reported in earlier papers [15,24,27,36]. In this study, we used the IVIM technique to quantify both conventional ADC values and tissue diffusion coefficient D t to better investigate the role of pure cellularity [38]. We found significantly lower ADC and D t values of PTs in HPV-positive patients than HPV-negative patients, which is consistent with the previous literature [24,27]. However, when evaluating the potential interaction between diffusion parameters and tumor subsite, it emerged that the latter had a significant impact on ADC and D t parameters of PTs, and only in the base of the tongue was the difference in D t between HPV-negative and HPV-positive patients on the significance threshold.
Compared to our previous work [15], a significant difference was also reported in D t and ADC for LNs in a larger population. Contrary to PTs, no differences in ADC and D t parameters of LNs were found in relation to the tumor subsite, suggesting that ADC/D t and tumor subsite may be considered independent predictors of HPV status; this is consistent with the results of the multivariate regression model, as discussed below.
The stronger statistical significance of D t , in comparison with ADC, suggest that it may be more convenient to eliminate the perfusion contribution from the diffusion coefficient estimation to more accurately investigate cellular microstructure. These differences in D t were also confirmed after dividing HPV-positive patients based on smoking and alcohol habits; in fact, it can be noticed that D t values of HPV-positive non-smoker non-alcoholic patients and HPV-positive smoker and alcoholic patients substantially overlap for both PTs and LNs.
It was hypothesized by Driessen et al. [39] that differences in diffusion parameters by HPV status may be connected to a smaller tumor-stromal component that is present in HPV-positive OPSCC at pathology and was correlated to ADC values, in agreement with subsequent studies [40]. In a study of Ward et al. [41], the separation in ADC estimates between HPV-negative and HPV-positive OPSCC was attributed to increased levels of lymphocytes in the latter, which may justify the decreased ADC values and the more favorable prognosis of this patient group.
The possible impact of HPV status on DCE-MRI parameters was also investigated. Consistent with earlier literature, any difference in vascularity was reported based on HPV status for primary tumors and lymph nodes [36,42,43]. The multivariate regression model had good predictive performance (AUC = 0.87) and indicated that the most significant factors associated with HPV status were tumor subsite, alcohol intake, and D t of LN. In particular, tumors located in the tonsil, compared to those located in the base of the tongue, were associated with a higher risk of HPV positivity according to a recent investigation of Bos et al. [44], while moderate or heavy alcoholic patients and increased D t values of lymphadenopathy were associated with a lower risk of HPV-positive OPSCC [45,46].
Some potential limitations of our study need to be acknowledged. Even though the patient population was quite large, there was a lower proportion of HPV-negative patients, which may have introduced a potential statistical sample bias. Moreover, further quantitative analyses could have been proposed, particularly on DWI, which has been demonstrated to be more informative in relation to HPV status [40]. Future investigations could also include radiomic analysis from D t maps to explore the diagnostic performance of predictive multifactorial models of HPV positivity using an independent external cohort of patients to validate the models.
In conclusion, our study suggests that the combination of clinical factors, such as tumor subsite and alcohol habits, with the perfusion-free diffusion coefficient D t of lymphadenopathy may help to accurately discriminate OPSCC by HPV status.
Supplementary Materials: The following are available online at https://www.mdpi.com/article/10 .3390/jcm10173865/s1, Table S1: ADC and D t coefficients in primary tumors (PTs) and metastatic lymph nodes (LNs) by the tumor subsite; Table S2: ADC and D t coefficients in primary tumors (PTs) sorted by the HPV status, in Tonsil and Base of the tongue subgroups, separately; Table S3: Comparison between imaging parameters of HPV-negative, HPV-positive non-smoker non-alcoholic patients, and HPV-positive smoker and alcoholic patients, for primary tumors (PTs) and lymph-nodes (LNs); Table S4: Results of Spearman's correlation tests between ADC/D t and volume size of primary tumor (PT) and lymph node (LN); Table S5: ADC and D t coefficients in lymph nodes (LNs) with and without extra-capsular spread (ECS).  Institutional Review Board Statement: This study was approved by our institutional review board (approval number RS716/15).
Informed Consent Statement: Specific, informed consent was obtained from each patient.