The Application Value Of MR Diffusion Model And Radiomics In Salivary Gland Tumor

Part 1The diagnostic value of radiomics analysis based on themonoexponential DWI model for epithelial salivary glandtumorsBackground:The pathological types of epithelial salivary gland tumors are complex and diverse.The biological behavior of different tumor types varies,which affects disease treatment planning and prognosis prediction.The preoperative differentiation of benign from malignant salivary gland tumors and tumor subtypes is important.Fine needle aspiration cytology(FNAC)is an important tool for the preoperative diagnosis of salivary gland tumors,but it may cause the diffusion of tumor cells and the recurrence of malignant tumors.And some benign lesions can be confused with malignant tumors in small and insufficient biopsy specimens,such as the differentiation of pleomorphic adenoma and mucoepidermoid carcinoma,basal cell adenoma and basal cell adenocarcinoma.Magnetic resonance imaging(MRI)has become the common method in preoperative diagnosis of salivary gland tumors.However,the diagnostic value of conventional imaging features,such as clear or unclear boundary,signal homogeneity or heterogeneity,and signal intensity,for different histopathologic subtypes remains controversial.In recent years,radiomics has been widely concerned in the field of clinical oncology,of which the main advantage is that it can extract high-throughput quantitative data from standard medical images by characterizing comprehensively intratumoral heterogeneity to develop diagnostic,predictive,or prognostic models for improving clinical treatment planning.Diffusion model is a widely used functional MRI technology,and Diffusion weighted imaging(DWI)technique is more mature.At present,it is not clear whether combining optimal machine-learning model and DW imaging data can provide additional information to the tumor heterogeneity for the differentiation of benign from malignant salivary gland tumors.This study explores whether radiomics analysis based on DWI can provide more tumor information than conventional DWI.Objective:This work aims to evaluate the diagnostic value of radiomics analysis based on the monoexponential DWI model for the differentiation of benign from malignant epithelial salivary gland tumors and tumor subtypes.Materials and methods:A total of 262 patients with histopathology-confirmed epithelial salivary gland tumors from January 2015 to March 2019 were enrolled in this study,including 100 pleomorphic adenomas,68 Warthin tumors,45 rare benign tumors and 49 malignant tumors.The data sets of benign and malignant salivary gland tumors were divided into training set(170 benign tumors,39 malignant tumors)and validation set(43 benign tumors,10 malignant tumors).The data sets of salivary gland tumor subtypes were divided into training set(54 Warthin tumors,80 pleomorphic adenomas,39 malignant tumors)and validation set(14 Warthin tumors,20 pleomorphic adenomas,10 malignant tumors).Conventional MRI and DWI scanning were performed on the same Siemens 3.0 T MR scanner,and DWI images were selected for radiomics analysis.All regions of interest(ROI)were drawn manually on the three largest successive slices of the tumor based on the high b-value DW images.The reproducibilities of ROI delineation were determined using inter-and intraclass correlation coefficients(ICC).A total of 396 quantitative radiomic features were calculated from each patient based on A.K.software.The radiomic features were divided into six groups:42 histogram,9 form factor,10 Haralick,11 gray-level size zone matrix(GLSZM),144 gray-level co-occurrence matrix(GLCM),and 180 run length matrix(RLM).ICC analysis was conducted in radiomic features to assess the reproducibility of each feature.Stable features were defined by ICC greater than 0.75.The analysis of variance(ANOVA)and least absolute shrinkage and selection operator(LASSO)regression were used to single out the most valuable radiomic features.Three supervised machine-learning algorithms,namely,logistic regression method(LR),support vector machine(SVM),and K-nearest neighbor(KNN)were calculated for the construction of the diagnosis model,which were verified in the validation set.The diagnostic performances of the three predictive classification models were evaluated by receiver operating characteristic(ROC)curve and area under the curve(AUC)in the training and validation data setsResults:1.Two-classification diagnosis model of benign and malignant salivary gland tumors:During the feature selection and classifier training,fourteen most valuable features were investigated using LASSO regression.LR and SVM methods exhibited good diagnostic ability to predict benign and malignant tumors.In the training data set,LR and SVM yielded AUC values of 0.81 and 0.80,respectively,however,KNN showed relatively lower AUC(0.64).In the testing data set,a similar result was found,where AUC values for LR,SVM,and KNN were 0.83,0.76,and 0.57,respectively.2.Three-classification diagnosis model of salivary gland tumor subtypes:During the feature selection and classifier training,twenty most valuable features were investigated using LASSO regression.In the training data set,LR yielded AUC values of 0.818,0.812,and 0.857 in predicting Warthin tumor,pleomorphic adenoma and malignant tumors respectively,and SVM yielded AUC values of 0.793,0.796,and 0.824,respectively and KNN yielded AUC values of 0.737,0.759 and 0.785,respectively.In the testing data set,the AUC values for LR were 0.712,0.706,and 0.932,respectively,those for SVM were 0.726,0.723,and 0.912,respectively,and those for KNN were 0.749,0.678,and 0.778,respectively.Conclusion:1.The radiomics models based on DW images present a certain predictive value in distinguishing benign and malignant epithelial salivary gland tumors.2.Three-classification model on DWI had good diagnostic efficiency for histological subtypes of salivary gland tumor(Warthin tumor,pleomorphic adenoma and malignant tumors).3.LASSO-LR and LASSO-SVM classifiers were relatively stable machine learning methods,which yielded the best diagnosis performance for predicting malignant tumors in both training and validating sets,so as to guide clinical decision-making.Part 2The diagnostic value of advanced IVIM-DKI model indifferentiating salivary gland tumorsBackground:MR diffusion model is an imaging technique to describe the water diffusivity of microstructure in the tissue.Diffusion weighted imaging(DWI)plays a leading role in the clinical application of head and neck tumor detection,staging,feature description and treatment response prediction.As an extension of monoexponential DWI technique,advanced diffusion models,such as intravoxel incoherent motion(IVIM)and diffusion kurtosis imaging(DKI),should be more sensitive to obtain functional information related to complex tissue microstructure,so as to provide more reliable quantitative features of the tumor.DWI and its derived apparent diffusion coefficient(ADC)have a high accuracy in the diagnosis of salivary gland tumors,while other studies have shown that ADC values of different histological types overlap.Thus,no consensus has been established with regard to the role of DWI in differentiating salivary gland tumors.At present,IVIM or DKI technique is initially used in the differential diagnosis of parotid diseases,and its diagnostic efficacy needs further study.The combination of multiple non-Gaussian diffusion models can comprehensively evaluate tumor cellularity,tissue heterogeneity,and perfusion,etc.Previous studies have confirmed that IVIM-DKI model can help to differentiate four histological types of sinonasal malignancies.Therefore,our study attempts to explore the diagnostic value of combing these two advanced diffusion models to distinguish different histological types of salivary gland tumors.Objective:To evaluate the diagnostic value of advanced IVIM-DKI model in discriminating benign and malignant salivary gland tumors and different tumor subtypes,in order to improve the ability of preoperative qualitative diagnosis.Materials and methods:From January 2018 to January 2020,patients with suspected salivary gland tumors before surgery were collected prospectively in lining First People's Hospital.The inclusion criteria were as follows:1)All patients underwent preoperative multi-parameters MRI scan on the same Philips 3.0T MRI equipment,including IVIM,DKI and dynamic enhanced scan sequence,etc.2)All patients did not have any invasive therapy before MR examination.3)Clinical and pathological data were complete.The exclusion criteria were as follows:1)Image artifacts and deformation caused by various reasons affected measurement.2)It was difficult to measure accurately the small lesion due to volume effect.3)The lesions with obvious cystic component can not be accurately measured.Finally,50 patients with salivary gland tumor confirmed by histopathology were included in the study.Image post-processing and data acquisition:All measurements of IVIM/DKI parameters were performed using IMAgenGINE MRI Diffusion/Perfusion Toolbox.Regions of interest(ROIs)were manually drawn in the maximum solid region of tumor avoiding obvious necrosis,cystic,and hemorrhage portions based on contrast-enhanced scanning images.The pure diffusion coefficient(D),pseudodiffusion coefficient(D*),perfusion fraction(f),mean kurtosis coefficient(MK)and mean diffusion coefficient(MD)were calculated for 3 times,and then averaged the results.The reproducibilities of inter-observer and intra-observer measurements were analyzed by inter-and intraclass correlation coefficients(ICC).Statistical Analysis:All statistical analyses were performed using SPSS 22.0 and MedCalc13.0.The mean±standard deviation was used to describe the numerical variables.The frequency was used to describe the categorical variables.Independent sample t-test or one-way ANOVA or rank sum test were used for comparing different quantitative or qualitative data.Bonferroni method was used to correct the p value for comparison between two histological types Receiver operating characteristic(ROC)curve analysis was performed to evaluate the diagnostic performances of different parameters,and the accuracy,sensitivity and specificity were calculated.The areas under the ROC curve(AUC)of different parameters were compared using the Delong et al.Method.For joint analysis of multiple parameters,we used the multi-factor logical regression analysis.In all statistical analysis,P<0.05 was considered statistically significant.Results:Fifty patients with salivary gland tumor included 50 lesions,of which 39 were benign and 11 were malignant.The pathological types of benign group included twenty-one pleomorphic adenomas(53.8%),eleven Warthin tumors(28.2%),six basal cell adenomas(15.4%),and one hemangioma(2.6%).The pathological types of malignant group included three acinic cell carcinomas(27.3%),one adenocarcinoma(9.1%),one invasive carcinoma(9.1%),one adenoid cystic carcinoma(9.1%),one secretory carcinoma(9.1%),one carcinoma in pleomorphic adenoma(9.1%),one metastatic squamous cell carcinoma(9.1%),one primary squamous cell carcinoma(9.1%),and one salivary duct cancer(9.1%).The D value and MD value of benign group were higher than those of malignant group(1.318±0.467×10-3mm2/s vs 1.007±0.099×10-3mm2/s,P=0.049;1.581±0.577×10-3mm2/s vs 1.151±0.155×10-3mm2/s,P=0.027),the MK value of benign group was lower than that of malignant group(0.747±0.302 vs 0.979±0.164,P=0.008).The AUCs of D value,MD value,and MK value for distinguishing benign and malignant salivary gland tumors were 0.696,0.718,and 0.758 respectively,with the accuracy of 78%,76%,and 74%respectively,with the specificity of 64.1%,59%,69.2%respectively,and with all sensitivity of 100%;By combining D value,MD value and MK value,the AUC increased slightly to 0.769.The D value and MD value of pleomorphic adenomas were higher than those of malignant group(1.595±0.427×10-3mm2/s vs 1.007±0.099×10-3mm2/s,P<0.001;1.923±0.525×10-3mm2/s vs 1.151±0.155×10-3mm2/s,P<0.001),and the MK value of pleomorphic adenomas was lower than that of malignant group(0.53 8±0.128 vs 0.979±0.164,P<0.001).The AUCs of D value,MD value,and MK value were 0.939,0.935,and 1.000 for distinguishing pleomorphic adenomas and malignant group respectively,with the accuracy of 87.5%,87.5%,and 100%respectively,with the sensitivity of 100%,90.9%,and 100%respectively,and with the specificity of 90.5%,90.5%,and 100%respectively.The D value and MD value of Warthin tumors were lower than those of malignant group(0.807±0.085×10-3mm2/s vs 1.007±0.099×10-3mm2/s,P<0.001;0.936±0.091×10-3mm2/s vs 1.151 10.155×10-3mm2/s,P=0.003),the D*value of Warthin tumors was higher than that of malignant group(59.134±11.454×10-3mm2/s vs 38.655±9.022×10-3mm3/s,p<0.001).The AUCs of D value,D*value,and MD value for distinguishing Warthin tumors and malignant group were 0.938,0.926,and 0.876 respectively,with the accuracy of 81.82%,77.27%,and 77.27%respectively,with the sensitivity of 100%,81.8%,and 72.7%respectively,and with the specificity of 72.7%,90.9%,and 100%respectively;By combining D value and D*value or combining MD value,D value and D*value,the AUCs had an upward trend of 0.967 and 0.983,respectively.The D value and MD value of pleomorphic adenomas were higher than those of Warthin tumors(1.595±0.427 ×10-3mm2/s vs 0.807±0.085 ×10-3mm2/s,P<0.001;1.923±0.525 ×10-3mm2/s vs 0.936±0.091 ×10-3mm2/s,P<0.001).The D*value and MK value of pleomorphic adenomas were lower than those of Warthin tumors(33.98±10.727 ×10-3mm2/s vs 59.134±11.454 ×10-3mm2/s,P<0.001;0.538±0.128 vs 1.166±0.107,P<0.001).The AUCs of D value,D*value,MD value,and MK value were 0.987,0.952,1.000,and 1.000 for distinguishing pleomorphic adenomas and Warthin tumors,with the accuracy of 96.87%,81.25%,100%,and 100%respectively,with all sensitivity of 100%,and with the specificity of 95.2%,81%,100%,and 100%respectively.Conclusion:1.D value,MD value and MK value may be helpful to distinguish benign and malignant salivary gland tumors.2.All D value,D*value,MD value,and MK value had high diagnostic efficiency in differentiating pleomorphic adenomas and Warthin tumors.3.All D value,MD value and MK value had high diagnostic efficiency in differentiating pleomorphic adenomas and malignant tumors.4.By combining D value and D*value or combining MD value,D value and D*value,the diagnostic efficiency was improved in distinguishing Warthin tumors from malignant tumors.