Texture Analysis Of Non-Enhanced CT Images:Differentiation Of Parotid Malignant Tumors And Common Benign Tumors

Objective : To evaluate the possibility of texture analysis on non-enhanced CT images for differentiating benign and malignant tumors of the parotid gland and several common benign tumors of the parotid gland(pleomorphic adenoma,adenolymphoma,basal cell adenoma).Materials and Methods: This retrospective study included 201 cases of parotid gland tumor confirmed by pathology(including 117 cases of pleomorphic adenoma,39 cases of adenolymphoma,19 cases of basal cell adenoma and 26 cases of primary malignant tumor of the parotid gland).Artificial Intelligent Kit texture analysis software was used to delineate region of interest(ROI)on the lesions of these non-enhanced CT images,and obtained 18 texture parameters.These texture parameters were compared by the Mann-Whitney U test in differentiating benign and malignant tumors of the parotid gland,and then compare the differences of texture parameters between the three types of parotid benign tumors in pairs.The ROC curve was used to evaluate the diagnostic effect of parameters with statistical difference.Results: There were 18 different CT texture parameters generated from these non-enhanced CT images,which were min intensity ? max intensity ? median intensity?mean value?standard deviation?variance?volume count?voxel value sum?range?root mean square?mean deviation?relative deviation?skewness?kurtosis ? uniformity ? histogram energy ? histogram entropy ? frequency size,respectively.Among them,9 texture parameters were statistically significant between the benign tumor group and the malignant tumor group(P<0.05),which were max intensity?median intensity?mean value?volume count?voxel value sum?root mean square?mean deviation?relative deviation?frequency size,respectively. Among them,volume count ? voxel value sum ? frequency size have higher diagnostic efficacies,the AUCs are 0.747?0.758?0.747,respectively.Among the three types of benign parotid tumors,13 parameters were statistically significant between the pleomorphic adenoma group and the adenolymphoma group(P<0.05),which were min intensity?max intensity?median intensity?mean value?volume count?voxel value sum?range?root mean square?mean deviation?uniformity?histogram energy?histogram entropy?frequency size,respectively.Among them,max intensity?median intensity?mean value?voxel value sum?mean deviation have higher diagnostic efficacies,the AUCs are 0.815 ? 0.815 ? 0.806 ? 0.815 ?0.806,respectively.There were statistically significant differences in 10 parameters between the adenolymphoma group and the basal cell adenoma group(P<0.05),which were min intensity?max intensity?median intensity?mean value?volume count?voxel value sum?root mean square?mean deviation?uniformity?frequency size,respectively.Among them,median intensity?mean value?root mean square?mean deviation have higher diagnostic efficacies,the AUCs are 0.820?0.812?0.814?0.812,respectively.There were no significant differences in the 18 texture parameters between the pleomorphic adenoma group and the basal cell adenoma group(P>0.05).Conclusion: Texture analysis of non-enhanced CT images can be used to identify benign and malignant tumors of the parotid gland,parotid pleomorphic adenoma and adenolymphoma,parotid adenolymphoma and basal cell adenoma.