Amide Proton Transfer-weighted MRI in the Diagnosis of Major Salivary Gland Tumors

Amide proton transfer-weighted magnetic resonance imaging (APTw-MRI), which is effective in tumor characterization, has expanded its role in the head and neck. We aimed to evaluate the diagnostic ability of APTw-MRI in differentiating malignant from benign major salivary gland tumors compared with diffusion-weighted imaging (DWI) and dynamic contrast-enhanced (DCE)-MRI. Between December 2017 and November 2018, 38 subjects, who were diagnosed with major salivary gland tumors and who underwent preoperative 3 T MRI, including APTw-MRI, DWI, and DCE-MRI, were included in this retrospective study. Twenty-three subjects had benign tumors, and fifteen had malignancies. APTw-signals of the tumors were measured and compared according to the histopathological diagnosis. Using receiver operating characteristic curve analysis, diagnostic performance of APTw-MRI was evaluated and compared with DWI and DCE-MRI using DeLong test. The maximum, mean, and median APTw-signals were significantly higher in malignant than in benign tumors (P < 0.001). The mean and maximum APTw-signals showed excellent area under the curve for predicting malignant tumors (0.948 and 0.939), which were significantly higher than the combining use of DWI and DCE-MRI (0.780) (P = 0.021 and 0.028). Therefore, APTw-MRI could be a useful tool for differentiating malignant from benign major salivary gland tumors, and can be applicable in the clinical setting.

In this study, we adopted APTw-MRI in the preoperative assessment of parotid and submandibular gland tumors. We believed that parotid and submandibular spaces could be eligible for APTw-MRI with tolerable field inhomogeneity. We sought to assess the diagnostic performance of APTw-MRI in differentiating malignant from benign salivary tumors, and compared it with multi-parametric analysis using DWI and/or DCE-MRI. The purpose of this study was to determine the utility of APTw-MRI in differentiating between benign and malignant major salivary gland tumors compared with DWI and/or DCE-MRI.

Results
Demographics. There was no significant difference in age of the patients between benign (median 56 years, range 18-72 years) and malignant (62 years, 38-78 years) tumors (P = 0.055). Also, sex of the patients did not differ significantly between benign (15 females and 8 males) and malignant (5 females and 10 males) tumors (P = 0.096).
When we compared ROC curves from APTw-MRI and DWI and/or DCE-MRI using DeLong test, the AUCs for predicting malignancy were significantly higher with the use of mean and maximum APTw-signals than with the use of multi-parametric analysis (P = 0.021 and 0.028, respectively) and ADC alone (P = 0.02 and 0.05, respectively) (Fig. 3). The estimated sensitivity and specificity at the optimal cut-off level of each value are summarized in Table 2.

Discussion
In this current study, we revealed that the APTw-signals were significantly higher in malignant tumors arising in the major salivary glands than in benign tumors. Using the mean and maximum APTw-signals, excellent diagnostic performance in the prediction of malignant tumors was achieved, which was significantly higher than the multi-parametric analysis using DWI and/or DCE-MRI.
Preoperative diagnosis of major salivary gland tumors is important in surgical planning 3,4 . Fine needle aspiration -a common diagnostic procedure -is sometimes inconclusive due to sampling error or inaccessible tumor location 3,22,23 . Thus, the preoperative imaging becomes crucial in the diagnosis of major salivary gland tumors. However, conventional MRI features, such as signal intensity or tumor margin, have resulted in poor diagnostic outcomes 3,6 . A use of single functional parameter, such as ADC from DWI or TIC from DCE-MRI, can help precise diagnosis 24,25 . However, breakthrough was made by multi-parametric analysis with simultaneous usage of ADC and TIC, which showed high diagnostic accuracy in differentiating parotid tumors 3,7 .
Nevertheless, the role of APTw-MRI in characterizing head and neck tumors has not been actively evaluated thus far. Until now, only a single study group published two preliminary results regarding the utility of APTw-MRI in the head and neck region 5,21 . In their most recent study 21 , the authors analyzed the APTw-signals in head and neck cancers, including nasopharyngeal undifferentiated carcinoma, squamous cell carcinoma, and lymphoma, and concluded that the mean and median APTw-signals of these malignancies were significantly higher than those of benign salivary gland tumors and normal tissues. This study well demonstrated that APTw-MRI could effectively identify head and neck cancers. However, the authors did not include any malignant tumors from the salivary gland, and they only compared 14 cases of benign salivary gland tumors with a heterogeneous group of head and neck malignancies, excluding those with salivary gland origin. Thus, the capability of APTw-MRI in differentiating malignant from benign salivary gland tumors remains to be elucidated.
In our study, we focused on the major salivary gland tumors, and found that APTw-signal values were significantly higher in malignant tumors in parotid and submandibular glands than in benign tumors. This result was in accordance with the previous studies that suggested APTw-signals in malignancies can increase higher than in benign tumors due to increased glandular tumor cells containing abundant mobile proteins/peptides, leading to an increased effect of chemical exchange saturation transfer 13,16,20,21,26 . In addition, skewness and kurtosis were significantly higher in malignant tumors than in benign tumors. This may reflect the underlying pathologic

Cut-off value Sensitivity (%) Specificity (%)
Mean  www.nature.com/scientificreports www.nature.com/scientificreports/ architecture of the malignant tumors, with a large number of pixels possessing high APTw-signals from malignant tissues. Moreover, APTw-MRI achieved excellent diagnostic performance for malignant tumors, even higher than multi-parametric analysis combining DWI and DCE-MRI ( Supplementary Figs 2 and 3). It is worth noting that APTw-MRI has additional merit compared with DCE-MRI in that it can avoid exogenous contrast agent. APTw-MRI also can reduce the steps of image acquisition and decision-making processes compared with multi-parametric analysis combining DWI and DCE-MRI. Therefore, we could suggest that APTw-MRI may further add value in the assessment of major salivary gland tumors with clinical usefulness.
Compared with the previous studies that used DWI and DCE-MRI 3,7 , the overall sensitivity and specificity for the malignant tumors were lower in our study (sensitivity, 86% versus 73.3%; specificity, 92-100% versus 82.6%). We presume that this is likely due to our small study population, especially small number of malignant tumors. Another reason might be the differences in the image protocol for DWI and DCE-MRI. Thus, further validation of the optimal protocol for DWI and DCE-MRI as well as APTw-MRI is warranted.
There were several limitations in our study. First, the absolute magnitude of the percentage change in the APTw-signal depends on the imaging parameters of the pulse sequence 28 . Thereby, the absolute numbers of the APTw-signal from our study may differ from those from the previous studies 5, 21 . However, when using the same imaging method, the MTR asym (3.5 ppm) should always be higher in malignant tumors than in benign tumors 28 . Therefore, despite the differences in the absolute values, APTw-signal can be utilized in the differentiation of the malignant salivary gland tumors in the clinical setting. Second, as aforementioned, the number of our study population was small, and the sample size of malignant tumors was small compared with that of benign tumors; also, the malignant tumors consisted of various histological subtypes. Resultantly, the variation of APTw-signal was high in the malignant tumor group. However, since salivary gland tumors -especially malignant tumors -are not common, the low disease prevalence is the fundamental limiting factor. Furthermore, despite the heterogeneous histological types, our results showed that APTw-signals of the malignant tumors were significantly higher than those of the benign tumors. Since the decision of malignant versus benign tumor is more important than the histopathological diagnosis in the preoperative imaging, our result is relevant to the clinical practice. We believe that future study with more homogeneous and larger number of malignant tumors will reveal the improvement in the diagnostic performance with less signal variation. Third, we could not statistically analyze the added value of APTw-signals on DWI and DCE-MRI, also due to the small sample size. Therefore, further study with larger number of cases -especially malignant cases -is necessary to verify our results. Fourth, we did not exclude small-sized tumors from the analysis. Previous studies using APTw-MRI 5,21 have excluded small-sized tumors to allow reliable signal measurements. However, when we measured the region-of-interest (ROI) areas of tumors, the ROIs ranged in size from 100.61 to 2034.89 mm 2 . We believe that this value was sufficient for the measurement of APTw-signals under in-plane resolution of 2 × 2.5 mm 2 . Therefore, this should have minimal effect on our study results. Fifth, the ROI allocation could be subjective depending on two different readers. However, we aimed to minimize the possible inter-reader variance in placing ROI with several strategies detailed in the methods. In addition, the inter-observer agreements of APTw-signal measurements between the two readers were confirmed to be excellent and/or good by ICC analysis. We believe that this result can reflect that the issue of subjective ROI drawing could be solved via systemized allocation.
In conclusion, APTw-signals were significantly higher in malignant tumors of the major salivary gland than in benign tumors. The diagnostic performance of APTw-MRI in predicting malignancy was excellent and superior to DWI and DCE-MRI. In addition, APTw-MRI is a technique that can generate APTw-signal without using exogenous contrast materials. Therefore, APTw-MRI could be a useful tool for discriminating malignant from benign tumors of the major salivary glands, and can be applicable in the clinical setting.

Methods
subjects. This retrospective study was approved by the institutional review board of our institution, and written informed consent was waived. Between December 2017 and November 2018, 164 subjects underwent head and neck MRI for the evaluation of clinically suspected major salivary gland tumors. The inclusion criteria were as follows: (a) initial diagnosis of parotid or submandibular gland tumors, (b) pathologically proven tumors by fine needle aspiration, biopsy, or surgical resection, and (c) available preoperative 3 T MRI, including APTw-MRI, DWI, and DCE-MRI. The exclusion criteria were as follows: (a) prior treatment history for head and neck tumors (n = 56), (b) no record of fine needle aspiration, biopsy, or surgery, or inconclusive pathologic results (n = 12), (c) final pathology not confirmed as salivary gland tumors (external auditory canal cancer, n = 1; first branchial cleft cyst, n = 1; IgG4-related disease, n = 1; intramuscular lipoma, n = 3; sebaceous adenoma, n = 1; veno-lymphatic malformation, n = 3), (d) lack of available APTw-MRI, DWI, or DCE-MRI (n = 42), and (e) inadequate MRI quality (n = 6). Demographic data were obtained via electronic medical record.
As a result, a total of 38 subjects (median age, 58 years; age range, 18-78 years; 25 females and 13 males) were finally included. Thirty-six subjects had tumors in the parotid gland, and two subjects had tumors that originated from the submandibular gland. Final histopathological diagnoses were pleomorphic adenoma (n = 16), Warthin tumor (n = 6), oncocytoma (n = 1), epithelial-myoepithelial carcinoma (n = 2), mucoepidermoid carcinoma (n = 7), salivary duct carcinoma (n = 2), secretory carcinoma (n = 1), squamous cell carcinoma (n = 2), and carcinoma ex pleomorphic adenoma (n = 1), which constituted 23 benign and 15 malignant tumors. Imaging protocol. MRI was performed using a 3 T instrument (Ingenia CX; Philips Healthcare, Best, the Netherlands) with a 32-channel sensitivity encoding head coil. Coronal T2-WI with fat suppression, axial T2-WI with and without fat suppression, axial T1-WI without fat suppression were obtained, followed by APTw-MRI, DWI, and DCE-MRI. The acquisition of APTw-MRI covered entire tumor portion with a reference to axial T2-WI. Lastly, axial, coronal, and sagittal post-contrast T1-WI scans with fat suppression were performed. The imaging protocols for APTw-MRI, DWI, and DCE-MRI are detailed in Supplementary Materials. www.nature.com/scientificreports www.nature.com/scientificreports/ Imaging processing of Aptw-MRI. APTw-MRI processing was performed using home-developed Matlab (MathWorks, Natick, MA, USA) program. Water frequency shift was corrected based on water-saturation shift referencing images 29 . Z-spectrum for each voxel was fitted by the 12 th order polynomial model at all offset frequencies. Then, the fitted curve was interpolated to a higher resolution of 1 Hz. The actual water resonance frequency was assumed to be at the lowest signal of the interpolated Z-spectrum. The water center frequency offset was measured as the displacement between the actual and ideal water resonance frequency of 0 Hz. The APT Z-spectrum at six frequency offsets was interpolated over the offset range and shifted using the estimated water center frequency offset 29 . Based on the final shift-corrected Z-spectrum, asymmetric magnetization transfer ratio (MTR asym ) analysis was performed with respect to water frequency 9 . For APTw-imaging, MTR asym at 3.5 ppm was calculated as follows: APTw-signal measurement. APTw-signals were measured in each ROI on APTw-MRI. The following values were automatically calculated according to the histogram-based analysis: mean, minimum, maximum, and median values of APTw-signals, skewness, and kurtosis. The values of the APTw-signals from the two readers were averaged and used for further analysis.
DWI and DCE-MRI analysis. First, the mean ADC in each ROI was measured on the ADC map from DWI. Second, on DCE-MRI, the average signal intensities within the ROI were plotted against time to construct TIC. TICs were then classified into the following four types: type A, time to peak >120 seconds; type B, time to peak ≤ 120 seconds, high wash-out ratio (≥30%); type C, time to peak ≤ 120 seconds, low wash-out ratio (<30%); and type D, flat 3,24 . According to the previous scheme using both ADC and TIC 3 , we categorized the lesions with a >type A TIC, b >type B TIC and ADC < 1.0 × 10 −3 mm 2 /sec, c > type C TIC and ADC ≥ 1.4 × 10 −3 mm 2 /sec, and d > type D TIC into benign tumors; otherwise, the tumors were categorized into malignancy. statistical analysis. Continuous variables were expressed as the median and range. Demographic data between benign and malignant tumors were compared using Chi-square and Mann-Whitney tests. Inter-observer agreement on APTw-signals between two readers was evaluated by ICC: greater than or equal to 0.75, excellent agreement; 0.60-0.74, good agreement; 0.40-0.59, fair agreement; and less than 0.40, poor agreement. Differences in APTw-signals, ADC values and TIC patterns between benign and malignant tumors were compared using Mann-Whitney test, Chi-square test, and linear association test. Diagnostic performances predicting malignant tumors using APTw-signals, ADC values, TIC patterns, and the results from multi-parametric analysis combining DWI and DCE-MRI, were evaluated via ROC curve analysis. The AUC values from each ROC curve analysis were compared using DeLong test 30 . P values of less than 0.05 were considered to indicate significant differences. All statistical analyses were performed using SPSS v. 22.0 (SPSS, Chicago, IL, USA) and MedCalc 17.9 (MedCalc, Mariakerke, Belgium).