Differentiation of Malignant Salivary Gland Tumors from Pleomorphic Adenomas and Warthin’s Tumors: Combined Diagnostic Value of Tumor Blood Flow and Apparent Diffusion Coecient by Histogram Analysis

We aimed to evaluate the usefulness of tumor blood ow (TBF) obtained by pseudocontinuous arterial spin labeling (pCASL) and apparent diffusion coecient (ADC) for differentiating salivary gland malignant tumors (MTs) 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 signicant differences in the mean, 50th, 75th, and 90th percentiles of TBF among the tumor types, in the mean TBFs (ml/100g/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 mm 2 /sec) 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 MT and PA differentiation and 0.855, 0.814, and 0.905 for MT and WT differentiation, respectively. The combination of TBF and ADC evaluated by histogram analysis helped differentiate salivary gland MTs from PAs and WTs.

Malignant salivary tumors demonstrate a range of biological behaviors. About 40% of MTs are indolent, especially in young adults [3]. The other 40% of MTs are aggressive, especially in the elderly [3]. Clinical indicators suggesting MTs are rapid growth rate, pain, facial nerve involvement, and cervical adenopathy. However, a slow growth rate of asymptomatic mass does not exclude MTs [3]. Therefore, it is important to differentiate MTs from benign salivary gland tumors, such as PAs and WTs. Fine-needle aspiration cytology is widely accepted as a reliable way to diagnose salivary gland tumors before surgical resection, but it is not appropriate for tumors located in deep areas and is an intrinsically invasive procedure [4].
Noninvasive magnetic resonance imaging (MRI) techniques may improve the diagnostic performance of salivary gland tumors regardless of tumor locations. However, conventional MRI cannot clearly distinguish between benign and malignant salivary gland tumors [5]. For instance, the apparent diffusion coe cient (ADC) obtained by diffusion-weighted imaging (DWI) reportedly provided useful information for the differentiation of WTs and PAs but remained inconclusive for differentiation of benign and malignant salivary gland tumors [6-8].
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 ow (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 e ciency in the differential diagnosis of salivary gland tumors. In addition, appropriate automated software needs to be developed so that these advanced applications can be adjusted to facilitate the work ow of radiologists and to objectively evaluate quantitative data, such as ADC and TBF. Therefore, we have developed a custom software application in MATLAB 2020a for evaluation of TBF and the ADC of salivary gland tumors by using histogram analysis. The purpose of this study was 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 nally 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 ve carcinoma ex pleomorphic adenomas, two acinic-cell 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.   .87 mL/100 g/min, p = 0.020) and signi cantly lower in MTs than in WTs (147.45 ± 51.63 mL/100 g/min, p < 0.001).
Comparison of diagnostic performance for TBF and ADC in differentiating MTs, PAs, and WTs.
Tables 4, 5, and 6 show the diagnostic performance of each parameter determined by the receiver operating characteristic (ROC) curve analysis. When differentiating MTs from PAs, the 10th and 25th percentiles of the ADC both had the best diagnostic performance out of all TBF and ADC parameters, with areas under the curve (AUCs) of 0.885 and 0.885, respectively, which is considered medium diagnostic performance. The best detected cutoff points were 1.15 × 10 − 3 mm 2 /sec and 1.26 × 10 − 3 mm 2 /sec, respectively, yielding sensitivity and speci city for both cutoff values of 73.3% and 93.8%, 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, which is considered high diagnostic performance. The best detected cutoff point was 0.79 × 10 − 3 mm 2 /sec, yielding a sensitivity and a speci city of 100.0% and 89.5%, respectively. Figure 4 summarizes the diagnostic performance of the parameters. TBF, ADC, and the combination of TBF and ADC showed medium to high diagnostic performances, with AUCs of 0.850, 0.885, and 0.950 for differentiating MTs from PAs, 0.855, 0.814, and 0.905 for differentiating MTs from WTs, and 0.968, 1.000, and 1.000 for differentiating PAs from WTs, respectively.
Interobserver agreement of TBF and ADC measurements. Table 7 shows the intraclass correlation coe cients (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 and ADC for differentiating MTs from PAs and WTs increased relative to the performance of each parameter alone. However, in differentiating PAs from WTs, the diagnostic performance of ADC alone showed perfect discrimination, and therefore, the value of adding the combination of ADC and TBF was low. To our best knowledge, this is the rst 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.
Kato et al. reported that qualitative analysis showed that TBF was signi cantly higher in WTs than PAs and MTs, but did not show a signi cant 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 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]. In addition, MP-EPISTAR used in the study of Kato et al. has a lower signal-to-noise ratio than that of pCASL [11]. Thus, the pCASL technique may be more suitable for imaging compared to the ASL sequence that Kato et al. used for differentiation among MTs, PAs, and WTs.
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 signi cantly 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.  11], which probably contributes to the highest ADC value among the three types of tumors in all parameters except for skewness and kurtosis in our study. In contrast, WTs showed the lowest ADC among all parameters except for skewness and kurtosis, which might re ect epithelial and lymphoid stromata with microscopic slit-like cysts lled with proteinous uid [2,6].
There were several limitations in this study. First, the study was conducted at a single institution with a relatively small number of subjects. Studies with a larger number of subjects would be required to con rm the e cacy of pCASL imaging and ADC mapping for evaluating salivary gland tumors. Second, 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.
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.

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 study procedures were conducted according to the principles of World Medical Association Declaration of Helsinki. We  where λ is the blood/tumor-tissue water partition coe cient (1.0 g/mL), and SI control and SI label are the time-averaged 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 e ciency (0.85), SI PD is the signal intensity of a proton density-weighted image, and τ is the label duration (1650 ms). The value ofλwas 1.0 mL/g. To calculate TBF, we used the same model and conditions as those used for calculating blood ow in the brain.
Image analysis.
Image analysis was performed by using a custom software application developed in MATLAB 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-certi ed neuroradiologists (F.T and R.K) reviewed all MRI sequences. First, we identi ed 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.