Texture analysis of conventional magnetic resonance imaging and diffusion-weighted imaging for distinguishing sinonasal non-Hodgkin’s lymphoma from squamous cell carcinoma

To evaluate the value of texture analysis (TA) of conventional magnetic resonance imaging (MRI) and diffusion-weighted imaging (DWI) in the differential diagnosis between sinonasal non-Hodgkin’s lymphoma (NHL) and squamous cell carcinoma (SCC). Forty-two patients with sinonasal SCC and 30 patients with NHL were retrospectively enrolled. TAs were performed on T2-weighted image (T2WI), apparent diffusion coefficient (ADC) and contrast-enhanced T1-weighted image (T1WI). Texture parameters, including mean value, skewness, kurtosis, entropy and uniformity were obtained and compared between sinonasal SCC and NHL groups. Receiver-operating characteristic (ROC) curves and logistic regression analyses were used to evaluate the diagnostic value and identify the independent TA parameters. The mean value and entropy of ADC, and mean value of contrast-enhanced T1WI were significantly lower in the sinonasal NHL group than those in the SCC group (all P < 0.05). ROC analysis indicated that the entropy of ADC had the best diagnostic performance (AUC 0.832; Sensitivity 0.95; Specificity 0.67; Cutoff value 6.522). Logistic regression analysis showed that the entropy of ADC (P = 0.002, OR = 26.990) was the independent parameter for differentiating sinonasal NHL from SCC. TA parameters of conventional MRI and DWI, particularly the entropy value of ADC, might be useful in the differentiating diagnosis between sinonasal NHL and SCC.


Introduction
Sinonasal non-Hodgkin's lymphoma (NHL) and squamous cell carcinoma (SCC) are the two most common sinonasal malignancies [1,2]. Surgery is the first-line treatment strategy for patients with sinonasal SCC, while combination of radiotherapy and chemotherapy is usually suggested for patients with sinonasal NHL [3]. Therefore, accurate differential diagnosis between NHL and SCC is crucial for establishing individual treatment regimens. Differential diagnosis based on the clinical symptoms is quite difficult, because of that these two kinds of tumors can have mimicking symptoms, such as localized pain, nasal obstruction, epistaxis, unilateral facial or nasal swelling, nasal discharge, and headache [4]. Thus, to find a simple and effective method for establishing an accurate differential diagnosis is urgently needed in clinical practice.
Preoperative computed tomography (CT) and magnetic resonance imaging (MRI) are commonly used for the noninvasive assessment of sinonasal tumors [5][6][7]. Sinonasal NHL and SCC often share similar imaging characteristics, thus result in the poor diagnostic accuracy of conventional structural images features [5]. Recently, several studies have demonstrated that the mean apparent diffusion coefficient Guo-Yi Su and Jun Liu contributed equally to this work.
(ADC), which allows the characterization of water molecules diffusion in tumor tissues, might be helpful in distinguishing sinonasal NHL from SCC [6][7][8][9]. However, the mean ADC value is the average value of whole regions of interest (ROIs) in tumor tissues, which does not reveal the heterogeneity of tumor tissues.
Texture analysis (TA) which characterizes an image by analyzing the distribution and relationship of pixel gray levels, has attracted more and more attention [10]. It provides more objective and quantitative information regarding tumor heterogeneity beyond visual inspection [10]. TA has demonstrated its superiority in differentiating and grading tumors, or predicting treatment response in various diseases [10][11][12]. However, to the best of our knowledge, no study has applied TA of MRI to differentiate sinonasal NHL from SCC until now.
Therefore, the purpose of our study is to evaluate the value of TA of conventional MRI and DWI in the differential diagnosis between sinonasal NHL and SCC.

Study population
This study was approved by our institutional review board, and informed consents were obtained from all patients before MRI examinations. We included the patients who: (1) had no previous history of biopsy, surgery or any other treatment; (2) underwent conventional MRI and DWI for pretreatment evaluation of sinonasal tumors; (3) had an adequate MRI quality without motion and susceptibility artifacts; (4) had a final pathological diagnosis of sinonasal SCC or NHL. Finally, a total of 42 patients (male: female = 22:20; mean age 61.4 ± 12.7 years) with sinonasal SCC and 30 patients (male: female = 16:14; mean age 59.9 ± 17.9 years) with sinonasal NHL (17 with NK/T lymphoma, 13 with diffuse large B cell lymphoma) were included.

Image analyses
MR images were analyzed by two radiologists (with 7 and 3 years of experience in head and neck radiology, respectively) who were blinded to the study design and pathological information. T2WI and ADC map were registered to the CE-T1WI. ROIs were manually drawn in dedicated image processing software (FireVoxel; CAI2R; New York University, NY) with a filtration−histogram approach [13]. Tumor ROIs were manually delineated around all the slices on T2WI, ADC, and CE-T1WI (Fig. 1). Three-dimensional (3D) volumes of interest (VOIs) were automatically constructed by summing ROIs drawn in each section. Large necrotic, cystic, or hemorrhagic areas were carefully avoided with reference to T2WI and CE-T1WI. Lastly, volumetric regions were isotropically resampled to the in-plane resolution (voxel size = 1 × 1 × 1 mm) using cubic interpolation to ensure the conservation of scales and directions when deriving the 3D features [14]. After the VOIs were placed, following quantitative texture parameters were automatically obtained: mean value, skewness, kurtosis, uniformity, energy, and entropy, which are based on the histogram analysis and the gray-level co-occurrence matrix (GLCM) method [15].

Statistical analyses
Numeric data were averaged over all patients and reported as mean ± standard deviation. The data normality was tested by using the Kolmogorov-Smirnov test. Fisher's exact test was applied to assess the difference of binary variables between two groups. Texture parameters were compared between two groups using unpaired Student's t test or Mann−Whitney U test as appropriate. Receiver-operating characteristic (ROC) curves were performed to assess the performance of each significant texture parameter in the differential diagnosis between sinonasal NHL and SCC. Area under the ROC curve (AUC), sensitivity, and specificity were calculated. All the significant TA parameters were stepped into a multivariate logistic regression analysis to identify the independent parameter for differentiating sinonasal NHL from SCC. Inter-reader agreement was evaluated using intra-class correlation coefficient (ICC) with 95% confidence intervals, and classified as excellent (ICC ≥ 0.81), good (ICC = 0.61-0.80), moderate (ICC = 0.41-0.60), and poor (ICC < 0.40). All statistical analyses were performed by using MedCalc (version 13.0; Mariakerke, Belgium) and SPSS (version 24.0; IBM Corp., Armonk, NY, USA). A two-sided P value less than 0.05 was considered to be statistically significant.

Results
There were no significant differences in patients' gender (P = 0.995) and age (P = 0.778) between sinonasal NHL and SCC groups. Excellent inter-reader agreements were achieved in the measurement of TA parameters (ICC, 0.819-0.855). Detailed TA parameters of T2WI, ADC, and CE-T1WI are summarized in Table 1. There was no significant difference in all TA parameters derived from T2WI between two groups (all P > 0.05). As to the TA parameters from ADC, the mean value and entropy of ADC were lower in the sinonasal NHL group than those in the SCC group (P = 0.024, P = 0.001, respectively). As to the TA parameters from CE-T1WI, the mean value of CE-T1WI were higher in the SCC group (P = 0.008) than that in the NHL group (Fig. 2).
The results of ROC analysis are shown in Table 2. The optimal TA parameter alone was the ADC entropy, which yielded an AUC, sensitivity and specificity of 0.832, 95.2% and 66.7%, respectively (Fig. 3). Logistic regression analysis showed that ADC entropy (P = 0.002, OR = 26.990) was the independent parameter for differentiating sinonasal NHL from SCC. Representative cases of sinonasal NHL and SCC are shown in Fig. 4.

Discussion
Owing to its close relationship with individual treatment strategy, the differential diagnosis between sinonasal NHL and SCC was always one important clinical and research topic. As to conventional CT and MRI features, previous studies reported that bony erosion, tumor necrosis and soft tissue invasion might be useful for differential diagnosis [6,7]. However, semi-quantitative assessment of imaging features was a subjective process, and the underlying  information beyond the imaging was not fully excavated. As a promising approach, TA can provide more objective and quantitative information regarding tumor heterogeneity beyond visual inspection [10]. Therefore, in present study, we aimed to evaluate the performance of TA of ADC map and conventional MRI in differentiating sinonasal NHL from SCC. As a result, we found significant differences in ADC mean value, ADC entropy, and CE-T1WI mean value between two groups. Meanwhile, ADC entropy might be the most promising parameter in the diagnosis of sinonasal NHL from SCC. In present study, we found that there was no significant difference in those texture parameters of T2WI, which was in line with the previous studies [6,7]. They also reported that there was no significant difference in the signal intensity on T2WI between two groups [6,7]. Our study found that sinonasal SCC showed significantly higher mean value of CE-T1WI than NHL. This result was in good agreement with previous studies using dynamic-contrast enhanced MRI, in which NHL showed hypovascular perfusion [16,17]. Xiao et al. also indicated that SCC had higher fD * values than NHL by using intravoxel incoherent motion (IVIM) DWI, and fD * could be regarded as a promising imaging biomarker for the characterization of tumor perfusion [18]. Wang et al. reported that intratumoral heterogeneity and architecture of sinonasal tumors, such as tumor angiogenesis, might be revealed on CE-T1WI [19]. These results indicate that texture parameters may improve the diagnostic utility of CE-T1WI in the differentiation between NHL and SCC.
Currently, DWI was commonly used for preoperative diagnosis and grade of sinonasal tumors in clinical practice [6][7][8][9][10]. In our study, we found that the mean value of ADC was significantly lower in NHL than in SCC, which was in line with previous studies [6][7][8][9][10]. However, mean ADC value ignored the tumor heterogeneity, which was an important characteristic of malignant tumors. Entropy was a crucial texture parameter associated with the intra-lesion heterogeneity [20]. It has been proved to be a useful index for differentiating and grading tumors, and predicting disease prognosis [20]. Our study showed that SCC showed higher entropy of ADC than NHL, which was consistent   . 3 Performance of significant texture analysis parameters for differential diagnosing sinonasal SCC from NHL with previous study [7]. Kim et al. indicated that SCC was more heterogenous than NHL, due to intratumoral necrosis and hemorrhage [7]. Moreover, RESOLVE technique was used for DWI scan in our study. When compared with the widely used single-shot echo-planar (SS-EPI) technique, RESOLVE showed superiority in reducing distortion and artifact, and improving the overall imaging quality [21]. High image quality might be an important cornerstone for subsequent image analysis. This study has several limitations should be noted. First, we did not correlate the TA parameters of studied sinonasal NHL and SCC with the corresponding pathologic evaluation. Second, besides DWI scan, no other functional MRI techniques were combined used for pretreatment assessment. Thirdly, further radiomics analyze were not performed in this study due to limited sample size. Future study integrating more study population, more functional MRI modalities, and also radiomics analysis would be more attractive.
Our preliminary study found that TA parameter of conventional MRI and DWI, especially the entropy of ADC, might be useful for preoperative differential diagnosis between sinonasal NHL and SCC.