The Value Of Texture Analysis In Differential Diagnosis Of Fibrous Soft Tissue Tumors

Objective:To explore the value of texture analysis based on unenhanced CT images in differential diagnosis of fibrous soft tissue tumors in the musculoskeletal system.Materials and Methods:Fifty-two patients were retrospectively analyzed in different categories of fibrous soft tissue tumors confirmed by surgical pathology(19 cases in benign group,17 cases in intermediate group and 16 cases in malignant group).Routine CT unenhanced scan was performed for all patients.CT images were imported into 3D Slicer software.Two imaging diagnostic physicians independently evaluated CT images in the absence of unknown pathological results and applied Segment Editor kit to outline the lesions.Then the Radiomics plug-in was applied to input the Region of interest(ROI)and extract texture parameter features such as entropy,skewness,kurtosis,uniformity,energy,mean,contrast,correlation.Differences in image analysis were resolved through discussions with senior radiologists until a consensus was reached.SPSS 23.0 was used to analyze the corresponding data.The intra-class correlation coefficient(ICC)was used to test the consistency of the parameter values measured by the two observers.If the consistency was good,the average value of the two values was used for subsequent statistical analysis.The Kolomogorov-Smirnov method was used to test the normality of all measurement data.If the data conform to normal distribution,independent sample t test was adopted.If the data were skewed,Mann-Whitney U test was used.The difference was statistically significant if p<0.05.The Receiver operating curve(ROC)was drawn for statistically significant parameter values,and the area under the curve(AUC)was calculated.The higher AUC,the better the diagnostic efficiency.According to the maximum Jordan index(Jordan index=sensitivity+specificity-1),the threshold was determined,and the sensitivity and specificity under the threshold were calculated.Results:There are good consistency and high reliability for the data of parameters of three types of fibrous soft tissue tumors measured by two doctors.There are obvious statistical differences(p<0.05)for entropy,kurtosis,mean and correlation have in the comparison of benign and intermediate tumors,while no statistical differences(p>0.05)for skewness,uniformity,energy and contrast.In comparison between benign group and malignant group,there are statistical significance(p<0.05)for entropy,skewness,energy,contrast and correlation,while no statistical significance(p>0.05)for kurtosis,homogeneity and mean.In the comparison between the intermediate group and the malignant group,there are significant statistical differences in homogeneity,energy and contrast(p<0.05),while no statistical differences in entropy,skewness,kurtosis,mean and correlation(p>0.05).Based on CT plain scan sequence images,when differentiating benign group from intermediate group of fibrous soft tissue tumors,the AUC of correlation is 0.861 and the sensitivity and specificity are 82.4%and 89.5%respectively if the threshold of correlation is set as 0.570.The AUC of mean is 0.684and the sensitivity and specificity are 94.1%and 42.1%respectively if the threshold of mean is set as 15.223.When differentiating benign group from malignant group of fibrous soft tissue tumors,the AUC of correlation is 0.829 and the sensitivity and specificity are 81.3%and 89.5%respectively if the threshold of correlation is set as0.505.The AUC of entropy is 0.740 and the sensitivity and specificity are 62.5%and100.0%respectively if the threshold of entropy is set as 2.134.When differentiating intermediate group from malignant group of fibrous soft tissue tumors,the AUC of energy is 0.805 and the sensitivity and specificity are 100.0%and 62.5%respectively if the threshold of energy is set as 7.170×10~5.The AUC of uniformity is 0.688 and the sensitivity and specificity are 100.0%and 37.5%respectively if the threshold of uniformity is set as 0.245.Conclusion:Multiple parameters of texture analysis based on CT unenhanced images can be used for differential diagnosis of different subtypes of fibrous soft tissue tumors to some extent.