Tumor blood flow and apparent diffusion coefficient histogram analysis for differentiating malignant salivary tumors from pleomorphic adenomas and Warthin’s tumors

We aimed to assess the combined diagnostic value of apparent diffusion coefficient (ADC) and tumor blood flow (TBF) obtained by pseudocontinuous arterial spin labeling (pCASL) for differentiating malignant tumors (MTs) in salivary glands from pleomorphic adenomas (PAs) and Warthin’s tumors (WTs). We used pCASL imaging and ADC map to evaluate 65 patients, including 16 with MT, 30 with PA, and 19 with WT. We evaluated all tumors by histogram analyses and compared various characteristics by one-way analysis of variance followed by Tukey post-hoc tests. Diagnostic performance was evaluated by receiver operating characteristic (ROC) curve analysis. There were significant differences in the mean, 50th, 75th, and 90th percentiles of TBF among the tumor types, in the mean TBFs (mL/100 g/min) between MTs (57.47 ± 35.14) and PAs (29.88 ± 22.53, p = 0.039) and between MTs and WTs (119.31 ± 50.11, p < 0.001), as well as in the mean ADCs (× 10−3 mm2/s) between MTs (1.08 ± 0.28) and PAs (1.60 ± 0.34, p < 0.001), but not in the mean ADCs between MTs and WTs (0.87 ± 0.23, p = 0.117). In the ROC curve analysis, the highest areas under the curves (AUCs) were achieved by the 10th and 25th percentiles of ADC (AUC = 0.885) for differentiating MTs from PAs and the 50th percentile of TBF (AUC = 0.855) for differentiating MTs from WTs. The AUCs of TBF, ADC, and combination of TBF and ADC were 0.850, 0.885, and 0.950 for MTs and PAs differentiation and 0.855, 0.814, and 0.905 for MTs and WTs differentiation, respectively. The combination of TBF and ADC evaluated by histogram analysis may help differentiate salivary gland MTs from PAs and WTs.

Recently, arterial spin labeling (ASL) techniques, such as pulsed ASL or pseudocontinuous ASL (pCASL), were introduced for clinical applications 9 . This method has been applied for noninvasive measurement of tumor blood flow (TBF) by using the magnetization of protons in arterial blood as an intrinsic tracer without an exogenous contrast agent 9 . There have been only a few reports on the usefulness of ASL for differentiating salivary gland tumors so far [10][11][12] . The use of multiparametric MRI, such as DWI and ASL, may help radiologists by increasing their efficiency in the differential diagnosis of salivary gland tumors. This is because this method may decrease unnecessary examinations and invasive procedures, such as biopsies. We aimed to assess the combined diagnostic value of ADC and TBF for differentiating MTs in salivary glands from PAs and WTs.

Results
A total of 65 subjects (age range, 11-86 years; mean 59 years; 34 males and 31 females) were finally included. There were 16 subjects with MTs, 30 with PAs, and 19 with WTs. The characteristics of patients are described in Table 1. The pathology of MTs was variable, including five carcinoma ex pleomorphic adenomas, two aciniccell carcinomas, two adenocarcinomas, two adenoid cystic carcinomas, two mucoepidermoid carcinomas, one basal-cell adenocarcinoma, one epithelial myoepithelial carcinoma, and one salivary-duct carcinoma. One patient with PAs and eight patients with WTs had multiple or bilateral tumors. Among these patients, only the largest one was assessed.
Comparison of the parameters for TBF and ADC between MTs, PAs, and WTs. Figures Tables 2 and 3 show the parameter measurements of TBF and ADC, respectively, in MTs, PAs, and WTs.
There was a significant difference in the mean ADCs between MTs (1.08 ± 0.28 × 10 −3 mm 2 /s) and PAs (1.60 ± 0.34 × 10 −3 mm 2 /s, p < 0.001) but not between MTs and WTs (0.87 ± 0.23 × 10 −3 mm 2 /s, p = 0.117). There were no ADC parameters that showed significant differences for all three combinations of tumor types (MT and PA, MT and WT, and PA and WT). When differentiating MTs from WTs, the 50th percentile of TBF had the best diagnostic performance out of all TBF and ADC, with an AUC of 0.855 (95% CI, 0.733-0.977, p < 0.001), which is considered medium diagnostic performance. The best detected cutoff point was 78.02 mL/100 g/min, yielding a sensitivity and a specificity of 84.2% and 75.0%, respectively.
When differentiating PAs from WTs, the 10th percentile of ADC had the best diagnostic performance out of all TBFs and ADCs, with an AUC of 0.984 (95% CI, 0.958-1.000, p < 0.001), which is considered high diagnostic performance. The best detected cutoff point was 0.79 × 10 −3 mm 2 /s, yielding a sensitivity and a specificity of 100.0% and 89.5%, respectively. Figure 4 summarizes the diagnostic performance of the parameters. In differentiating MTs from PAs, the AUC for the combination of TBF all and ADC all (0.950; 95% CI, 0.892-1.000, p < 0.001) was higher than those for TBF all alone (0.850; 95% CI, 0.739-0.961, p < 0.001) and ADC all alone (0.885; 95% CI, 0.787-0.984, p < 0.001), which suggests that the diagnostic performance improved from medium to high with the combination of TBF all and ADC all . In differentiating MTs from WTs, the AUC for the combination of TBF all and ADC all (0.905; 95% CI, 0.805-1.000, p < 0.001) was higher than those for TBF all alone (0.855; 95% CI, 0.733-0.977, p < 0.001) and ADC all alone (0.814; 95% CI, 0.664-0.964, p = 0.002), which suggests that the diagnostic performance improved from medium to high with the combination of TBF all and ADC all . In differentiating PAs from WTs, the AUC for the combination of TBF all and ADC all (1.000; 95% CI, 1.000-1.000, p < 0.001) was higher than that for TBF all showing medium TBF (arrow). The region of interest (ROI) was manually drawn on the apparent diffusion coefficient (ADC) map of the software (e, yellow), and the ROI was copied from the ADC map to the TBF map of the software (d, yellow). The TBF histogram (f) and ADC histogram (g) are presented. The 50th percentile of the TBF value was 50.92 mL/100 g/min, whereas the 10th percentile of the ADC value was 0.82 × 10 −3 mm 2 /s. www.nature.com/scientificreports/ alone (0.968; 95% CI, 0.929-1.000, p < 0.001) and the same as that for ADC all alone (1.000; 95% CI, 1.000-1.000, p < 0.001), which suggested a medium diagnostic performance for TBF all alone and high performance for both ADC all alone and the combination of TBF all and ADC all . In differentiating MTs from benign tumors (BTs), including PAs and WTs, the AUC for the combination of TBF all and ADC all (0.930; 95% CI, 0.865-0.995, p < 0.001) was higher than those for TBF all alone (0.811; 95% CI, 0.709-0.914, p < 0.001) and ADC alone (0.895; 95% CI, 0.821-0.970, p < 0.001), which suggests that the diagnostic performance improved from medium to high with the combination of TBF all and ADC all . Supplementary Figure S1 presents the scatter plots for MTs, PAs, and WTs, which represent the propensity scores of TBF all and ADC all for each tumor. Table S5 shows the intraclass correlation coefficients (ICCs) of the measurements by the two observers. Excellent agreements were observed for all parameters except for the skewness of ADC, which showed good agreement.

Discussion
In this study, the diagnostic performance of the combination of TBF all and ADC all for differentiating MTs from PAs and WTs improved relative to the performance of each parameter alone. However, in differentiating PAs from WTs, the diagnostic performance of ADC all alone showed perfect discrimination, and therefore, the value of adding the combination of ADC all and TBF all was low. To our best knowledge, this is the first study to evaluate the usefulness of the combination of pCASL and the ADC map by histogram analysis for differentiating malignant salivary gland tumors from PAs and WTs.
According to Kato et al., qualitative analysis showed that TBF was significantly higher in WTs than PAs and MTs but did not show a significant difference between PAs and MTs 10 . However, we demonstrated that the mean, 50th, 75th, and 90th percentiles of TBF could differentiate MTs, PAs, and WTs. We speculate that the differences in ASL methods may explain why their results differed from ours. They placed the regions of interest (ROIs) on both a tumor and the contralateral normal parotid gland parenchyma at the same level and then evaluated showing a little heterogeneous contrast enhancement (arrow). Tumor blood flow (TBF) color map (c) showing low TBF (arrow). The region of interest (ROI) was manually drawn on the apparent diffusion coefficient (ADC) map of the software (e, yellow), and the ROI was copied from the ADC map to the TBF map of the software (d, yellow). The TBF histogram (f) and ADC histogram (g) are presented. The 50th percentile of the TBF value was 11.17 mL/100 g/min, whereas the 10th percentile of the ADC value was 1.71 × 10 −3 mm 2 /s. www.nature.com/scientificreports/ tumor-to-parotid signal intensity ratios from ASL images supposing that those ratios are surrogates of TBF 10 . They measured the relative ratio of salivary gland tumors to normal parotid glands, whereas we measured the TBF values of tumors quantitatively. Consequently, histogram analysis may overcome the limitations of qualitative analysis. Moreover, they used an alternating radio-frequency ASL sequence with gradient echo-type single-shot echo-planar imaging (MP-EPISTAR), which suffers from susceptibility artifacts more seriously than pCASL sequences that use 3D turbo spin-echo (TSE) acquisition 10    www.nature.com/scientificreports/ A recent report stated that metrics, such as percentiles, kurtosis, and skewness, calculated by histogram analysis are strong and reliable quantitative surrogate markers of tumor heterogeneity 13 . Thus, we consider that microenvironments of tumors could be masked by evaluating only a single parameter, such as the mean value. Yamamoto et al. demonstrated that the mean TBF value was significantly higher in WTs than in PAs by using the pCASL sequence with conventional ROI analysis 11 . They also showed that the higher mean TBF of WTs than of PAs was attributable to higher micro-vessel density in WTs than in PAs 11 . Furthermore, our results revealed that the 75th and 90th percentiles of TBF exhibited higher AUC values than the mean TBF. Consequently, histogram analysis appears to provide more detailed information about TBF.
Kato et al. reported that the mean ADC values were significantly higher in PAs than in WTs and MTs but were not significantly different between WTs and MTs 10 . Their results were consistent with our results showing that all ADC parameters except for skewness and kurtosis were significantly different between PAs and WTs and between PAs and MTs, but not between WTs and MTs. Razek et al. studied ADC values by histogram analysis for diagnosis of PAs, WTs, and MTs and reported significant differences in the means and skewness of ADC among all three tumors, although these differences between WTs and MTs were weaker than those between PAs and WTs and PAs and MTs 14 . Histopathologically, PAs comprise an abundant myxoma-like stroma 6,11 , which probably contributed to the highest value obtained for it among the three types of tumors in all ADC parameters, except for the skewness and kurtosis values for ADC, in our study. In contrast, WTs showed the lowest value among all ADC parameters, except for the skewness and kurtosis values for ADC, which might reflect epithelial and lymphoid stromata with microscopic slit-like cysts filled with proteinous fluid 2,6 .
Regarding the other conventional method, time-intensity curve patterns on dynamic contrast-enhanced MRI were found useful in the differentiation of salivary gland tumors 15 . Nevertheless, it requires contrast media, which can be harmful to patients with renal dysfunction or allergies to these materials. Moreover, dynamic contrastenhanced MRI only allows for one series of scans. In contrast, ASL can overcome these drawbacks and allows for repeat scanning without any contrast agents. Further, the time-intensity curve cannot provide quantitative data. For that reason, we focused on the noninvasive and quantitative MRI techniques of pCASL and ADC.
There were several limitations in this study. First, the study was conducted at a single institution with a relatively small number of subjects. Further studies with a larger number of subjects and a wider range of benign and malignant tumor types are required to confirm the efficacy of pCASL imaging and ADC map in evaluating salivary gland tumors. Furthermore, we should consider classifying malignant tumors into low, intermediate, and high grades and evaluate each group to facilitate the management of patients at an earlier stage. Regarding the analytical method, we could not evaluate the whole pCASL image slices and ADC maps for each tumor. Particularly, MTs tend to have heterogenous characteristics. Thus, whole-tumor evaluation would be desirable in future studies. Moreover, we evaluated limited parameters in histogram analysis. Thus, we need to consider other parameters, such as entropy, to provide further information on tumor heterogeneity.
In conclusion, the combination of TBF and ADC evaluated by histogram analysis was found to be helpful for differentiating MTs from PAs and WTs in salivary glands.

Methods
Subjects. This study was approved by the ethics committee of our university, and the requirement for written informed consent was waived because of the retrospective study design. All procedures were conducted according to the principles of the World Medical Association Declaration of Helsinki. We retrospectively identified 170 patients suspected of salivary gland tumors who had undergone pretreatment MRI between December 2015 and September 2020. Patients who fulfilled the following criteria were included: (a) available preoperative 3 T MRI with sufficient image quality, including pCASL images, DWI, T1-weighted images, contrast-enhanced T1-weighted images, and T2-weighted images; (b) tumor size > 10 mm; (c) tumors pathologically proven using fine-needle aspiration biopsy or surgical resection; and (d) diagnosis of MT, PA, or WT of the salivary gland. www.nature.com/scientificreports/ Patients were excluded on the absence of definitive diagnosis from biopsy or surgical resection (n = 37); histological diagnosis other than MT, PA, or WT (n = 19); lack of contrast-enhanced T1-weighted images (n = 20); lesions with large necrosis, cysts, hemorrhage, or infectious complications (n = 11); tumors smaller than 10 mm (n = 2); ADC map with artifact (n = 1); patients using a different pCASL protocol (n = 3); and data loading error in software (n = 12). A total of 65 patients met our inclusion criteria.   where λ is the blood/tumor-tissue water partition coefficient (1.0 g/mL), and SI control and SI label are the timeaveraged signal intensities in the control and label images, respectively. T 1,blood is the longitudinal relaxation time of blood (1650 ms), α is the labeling efficiency (0.85), SI PD is the signal intensity of a proton density-weighted image, and τ is the label duration (1650 ms). The value of λ was 1.0 mL/g. To calculate TBF, we used the same model and conditions as those used for calculating blood flow in the brain.

Conventional MRI protocol. All patients underwent MRI on a 3 T MRI system (Ingenia
Image analysis. Image analysis was performed by using a custom software application developed in MAT-LAB 2020a. The custom software displays the ADC map and the pCASL map for the same patient side by side on the monitor. A slice image of each map for display can be moved. Two board-certified neuroradiologists (F.T and R.K) reviewed all MRI sequences. First, we identified the tumors on T1-weighted images, T2-weighted images, and contrast-enhanced T1-weighted images. The ROIs were manually drawn around the tumor margin in the maximum diameters on the ADC map by using the software. The ROIs were within an entire solid part of a tumor as much as visually traced, avoiding areas of necrosis, cyst, or hemorrhage. Then, the segmented ROI was copied from the ADC map and pasted to the pCASL image by using the software. The histogram features for each image were determined using those histograms. The following 10 objective features were determined as histogram features in the custom software: (1) minimum (min), (2) mean, (3) maximum (max), (4) 10th percentile, (5) 25th percentile, (6) 50th percentile, (7) 75th percentile, (8) 90th percentile, (9) skewness, and (10) kurtosis. The histogram features of TBF and ADC were measured twice in each ROI, and these measurements were averaged.
Statistical analysis. Statistical analysis was performed by using SPSS v. 25.0 software (IBM SPSS Statistics for Windows, IBM Corp., Armonk, NY). Pearson's chi-square test was utilized to assess comparison of sex, tumor sub-site, and diagnostic method among the tumor types, and one-way analysis of variance was utilized to assess comparison of age and tumor diameter among the tumor types. All 10 parameters of the TBF and ADC values were assessed. Significant differences among the groups were analyzed using one-way analysis of variance followed by Tukey post-hoc tests, after Shapiro-Wilk test, which was performed to assess the normality of data distribution. A p value of < 0.05 was considered to be indicative of statistical significance. ROC curve analyses were performed to investigate the diagnostic performance of each parameter of TBF and ADC. All TBF parameters combined using binominal logistic regression were indicated as TBF all ; all ADC parameters combined using binominal logistic regression were indicated as ADC all ; and all TBF and ADC parameters combined using binominal logistic regression were indicated as TBF all + ADC all . These terms were used in differentiating MTs from PAs, MTs from WTs, PAs from WTs, and MTs from BTs, including PAs and WTs. We considered AUC values < 0.7, 0.7-0.9, and > 0.9 to indicate low, medium, and high diagnostic performance, respectively. Cutoff values were calculated with the maximum of the Youden index (Youden index = sensitivity + specificity − 1). A p value of < 0.05 was considered significant to be indicative of statistical significance.
Interobserver agreement on TBF and ADC values between two readers was evaluated by ICC. ICCs are considered excellent if > 0.74 16 . Ethics statement. This study was approved by the ethics committee of Mie University School of Medicine, and the requirement for written informed consent was waived because of the retrospective study design. All study procedures were conducted according to the principles of the World Medical Association Declaration of Helsinki.

Data availability
The datasets generated and analyzed during the current study are available from the corresponding author on reasonable request. TBF = 6000 · · (SI control − SI label ) · e PLD T 1, blood 2 · α · T 1,blood · SI PD · 1 − e