Classification Of Gastric Stromal Tumor

Objective:The purpose of this study was to analyze the image heterogeneity of gastric stromal tumor(GST)by computer tomography histogram analysis and to provide new objective quantitative imaging parameters for the prediction of grade GST.Methods : Totally,60 cases of GSTs with 62 tumors was enrolled in current retrospective analysis.According to the pathology,all the tumors were divided into four grades.Histogram parameters were derived using imaging software(Ommi-Kinetics,Version V2.1.0 R).Eighty one histogram features(27 parameters per phase)were selected and Kruskale Wallis was performed for the features selected.The histogram-feature values between benign(Grade I-III)and malignant(Grade IV)GSTs were compared by Logistic regression analysis,and receiver operating characteristic curve(ROC)was used to evaluate the efficacy of using histogram features to predict benign or malignant GSTs.Results:At quantitative evaluation,the best predictive parameters in differentiating between benign and malignant GSTs were Relative Deviation,Uniformity,Min Intensity and Volume Count.The area under ROC curve(AUC)showed that uniformity(artery phase)combined with Relative Deviation(delayed phase)(AUC = 0.935;p < 0.001)was of great significance in differentiating benign and malignant GSTs.Conclusion: This retrospective study highlighted the potential of CT histogram analysis to be a diagnostic marker in classification of GST.