CT-based Texture Analysis Study For Diagnosing Gastric Submucosal Tumors

Part One CT-based texture analysis study for differentiating gastric stromal tumors frombenign gastric stromal tumors Objective:This part is aimed at investigating the clinical value of CT-based texture analysis in differentiating gastric stromal tumors(GSTs)from benign gastric submucosal tumors(benign-GSMTs).Materials and Methods:The clinical,pathological and imaging data of 98 GSTs and 44 benign-GSMTs patients confirmed by pathology and underwent preoperative enhanced examination from August 2014 to August 2019 were analyzed retrospectively.The CT image features including location,growth pattern,shape,border,homogeneity,calcification,ulceration,necrosis and enlarged lymph nodes were evaluated independently.The longest diameter of primary lesions was measured.Meanwhile,the clinicopathological features were recorded.Kappa statistics was used to evaluate interobserver agreements for the assessments and measurements.The Student’s t-test(or Mann-Whitney U test)and the Chi-square test(or Fisher’s exact test)were applied to compare the differences of quantitative data and qualitative data,respectively.Logistic regression analysis was used to construct a clinical imaging model.The diagnostic performance of CT features was evaluated by ROC curve.The original image of the largest axial plane of the tumor in venous phase was imported into Mazda software in DICOM format for texture analysis.Drew the edge of the tumor,then ROI was obtained.When sketching,it should be close to the edge of the tumor to avoid the fat,air and water around the focus.After ROI was selected,the software would automatically generate the relevant parameters,and then extract the feature parameters.The Student’s t-test(or Mann Whitney U test)was used to compare the texture analysis parameters between the two groups.The logistic regression analysis was carried out in the single factor analysis with statistical significance,and the independent influencing factors to distinguish GSTs and benign-GSMTs were screened out.According to AUC value of ROC curve,the diagnostic value of CT image features and CT texture features for GSTs and benign-GSMTs was evaluated.Results:A total of 142 patients with gastric submucosal tumor were included in this study,including 98 cases of gastric stromal tumor and 44 cases of other gastric submucosal tumor(33 cases of gastric leiomyoma and 11 cases of gastric schwannoma).The results showed that age,tumor diameter,sex,growth pattern,tumor margin,enhanced homogeneity and necrosis could be used as independent factors to distinguish gastric stromal tumor and other submucosal tumors(P < 0.05).Kurtosis and entropy could be used as independent factors to distinguish gastric stromal tumor and other submucosal tumors(P < 0.05).ROC curve analysis showed that the AUC value of CT image model and image combined texture model were 0.852 and 0.932 respectively,the sensitivity were 72.7% and 86.4%,the specificity were 84.7% and 89.8%,respectively.The AUC,sensitivity and specificity of the combined texture model are higher than those of the CT model.Conclusion:1.The CT findings including age,tumor maximum diameter,gender,growth pattern,tumor margin,enhanced homogeneity and necrosis can be regarded as valuable factors for differentiating GSTs and benign-GSMTs.2.CT image combined with texture model has an important application value in the differential diagnosis of GSTs and benign-GSMTs,and its diagnostic performance is more excellent than that of CT image model,which is helpful for clinical diagnosis.Part Two CT-based texture analysis study for evaluating risk classification of gastric stromal tumorsObjective: This part is aimed at investigating the diagnostic values of CT-based texture analysis to determine the risk classification of gastric stromal tumors(GSTs).Materials and Methods: A total of 98 GSTs patients confirmed by pathology from August 2014 to August 2019 were retrospectively analyzed,with complete clinicopathological data and preoperative dual-phases enhanced CT scanning of abdominal in our hospital.The CT image features including location,growth pattern,shape,border,homogeneity,calcification,ulceration,necrosis and enlarged lymph nodes were evaluated independently.The longest diameter of primary lesions was measured.Meanwhile,the clinicopathological features were recorded.Intraclass correlation coefficient(ICC)was used to evaluate interobserver agreements for the assessments and measurements.The Student’s t-test(or Mann-Whitney U test)and the Chi-square test(or Fisher’s exact test)were applied to compare the differences of quantitative data and qualitative data,respectively.Logistic regression analysis was carried out on the CT image features with statistical significance in single factor analysis to screen out the independent influencing factors for distinguishing gastric stromal tumors with different risk classification.ROC curve was used to evaluate the diagnostic performance of CT images in different risk classification of gastric stromal tumors.The original image of the largest axial plane of the tumor in venous phase was imported into Mazda software in DICOM format for texture analysis.Draw the edge of the tumor,then ROI was obtained.When sketching,it should be close to the edge of the tumor to avoid the fat,air and water around the focus.After ROI was selected,the software would automatically generate the relevant parameters,and then extract the feature parameters.The texture analysis parameters between the two groups were compared by the Student’s t-test(or Mann Whitney U test).Logistic regression analysis was carried out on CT texture parameters with statistical significance in single factor analysis to screen out independent influencing factors for distinguishing gastric stromal tumors with different risk levels.The ROC curve was used to evaluate the diagnostic performance of CT image features combined with CT texture parameters in the evaluation of gastric stromal tumors with different risk levels.According to the AUC value of ROC curve,the diagnostic value of CT image features and CT texture features for low and high malignant gastric stromal tumors was evaluated.Results: In this study,98 cases of GSTs were included,including 58 cases of gastric stromal tumor in low-risk group and 40 cases of gastric stromal tumor in high-risk group.The results showed that the largest diameter,location,growth mode,shape,tumor margin,enhanced homogeneity,calcification and necrosis of tumors in CT images could be used as independent factors to distinguish the low-risk group fromthe high-risk group(P < 0.05).Mean,perc.10%,perc.50%,perc.90% and perc.99% of CT features could be used as independent factors to distinguish gastric stromal tumor in low-risk group and high-risk group(P < 0.05).ROC curve analysis showed that the AUC value of CT image model and image combined texture model were 0.942 and 0.964,respectively;the sensitivity were 82.5% and 85.0%,respectively;the specificity were 98.3% and 94.8%,respectively.The AUC,sensitivity and specificity of the combined texture model are higher than those of the CT model.Conclusion: 1.In the CT image diagnosis,the tumor’s maximum diameter,location,growth mode,shape,tumor margin,enhanced homogeneity,calcification and necrosis have certain significance in the differential diagnosis of gastric stromal tumors in low-risk and high-risk groups.2.CT image model and image joint texture model was constructed to distinguish gastric stromal tumors with different risk.The diagnostic performance of image joint texture model is more excellent.