| Objective:Radiomics model of the differential diagnosis of gastroin-testinal schwannoma(GS)and gastrointestinal stromal tumor(GIST)with different risk grades was established based on enhanced CT imaging,and the value of it was evaluated.Methods:Twenty-six patients with GS and 82 patients with GIST confirmed by pathology after surgery were collected in our hospital from August 2015 to November 2021.All enrolled patients did not receive any antitumor therapy and underwent enhanced abdominal CT examination before surgery.According to the postoperative pathology and the risk classification standard of central gastric stromal tumor according to the 2020 Chinese Society of Clinical Oncology CSCO Guidelines for the Diagnosis and Treatment of Gastrointestinal stromal Tumor,among the 82 GIST patients in this group,there were 7 very low risk,22 low risk,26 medium risk and 27 high risk GIST patients.The patients with very low risk and low risk were divided into 29 patients in the low malignant potential GIST group.Fifty-three patients were divided into the GIST group with high malignant potential.The enrolled patients were divided into training group and testing group in a ratio of 7:3.Clinical features(age,gender,gastrointestinal bleeding)of patients were recorded,and CT image features of tumors were analyzed,including:Lesion location(gastric fundus,gastric body,gastric antrum),the length to diameter,shape(regular or irregular),growing methods(inside,outside the cavity,inside and outside the cavity),the presence of tumor hemorrhage,presence of ulceration,and presence of cystic change,presence of liquefaction necrosis,presence of calcification,density is uniform,the boundary is clear,CT value,enhanced scanning arterial phase CT values,venous phase CT values,CT values at arterial stage,CT values at venous stage,pattern of tumor enhancement(uniform or uneven),and maximum peritumoral lymph node diameter.ITK-SNAP software was used to manually segment the whole tumor of thin layer images in the venous stage to extract the radiomics features.Feature screening was carried out through univariate analysis,correlation analysis,LASSO regression and multi-factor stepwise regression,and finally the radiomics features with independent influencing factors were retained.The traditional model was established based on clinical features and CT image features,the radiomics model was established based on the radiomics features,and the combined model was established based on the above features for the identification of GS and GIST,to evaluate the diagnostic efficacy of each model and its diagnostic efficacy in differentiating GS from GIST with different risk grades.Results:In the training group,there were 77 cases,19 cases of GS and 58 cases of GIST,including 21 cases of low malignant potential GIST and 37 cases of high malignant potential GIST.In the testing group,there were 31 cases,7 cases of GS and 24 cases of GIST,including 8 cases of low malignant potential GIST and 16 cases of high malignant potential GIST.(1)There were significant differences in tumor location,length diameter,cystic degeneration,liquefaction and necrosis,uniform density,venous CT value,net increase of arterial CT value,and uniform enhancement between GS and GIST in the training group(r<0.05).The location of the tumor,the presence or absence of cystic lesion,the net increase of arterial CT value and the uniform enhancement were independent predictors of distinguishing GS from GIST.Based on this,the traditional model was established,and the AUC of the training group and the validation group were 0.939 and 0.869,respectively.The accuracy rates were 0.818 and 0.774,respectively.(2)A total of 1595 radiomics features were extracted,and 8 radiomics features were retained to construct the logistic regression model.The AUC of the training group and the testing group were 0.949 and 0.839,respectively,and the accuracy rates were 0.922 and 0.774,respectively.(3)The traditional features of tumor location,cystic degeneration,arterial CT value increase,uniform enhancement,and radiomics model output(rad-score)were used for multi-factor stepwise regression.Tumor location,uniform enhancement,net increase in arterial CT value and rad-score were independent predictors for differentiating GS from GIST.Based on this,the combined model was constructed,and the AUC of the training group and the testing group were 0.989 and 0.964,respectively.The accuracy rates were 0.961 and 0.871,respectively.(4)The Integrated Discrimination Improvement(IDI)was used to compare the diagnostic efficiency of traditional model,radiomics model and combined model.In the efficiency comparison between the combined model and the traditional model,IDI value was 0.2538(r=0.00023,<0.05),indicating that the diagnostic efficiency of the combined model was a positive improvement over the traditional model,and the difference between the two was statistically significant.The DCA curves of the three models were all higher than the two reference lines,and the clinical net benefit of the combined model was higher than the traditional model.(5)In the training group and testing group,the diagnostic efficiency of the traditional model,the radiomics model and the combined model in differentiating GS from GIST with high malignant potential was much higher than that of GS from GIST with low malignant potential.The diagnostic efficacy of combined model in distinguishing GS from GIST with high malignant potential and GS from GIST with low malignant potential was higher than that of traditional model(IDI=0.2418,r<0.05;IDI=0.2749,r<0.05).Conclusions:1.Tumor location,net increase in arterial phase CT value,uniform enhancement,and rad-score were independent predictors for distinguishing GS from GIST.2.The diagnostic efficacy of the combined model based on CT image features and radiomics features in the identification of GS and GIST with different risk grades is higher than that of the traditional model,and it may become a new non-invasive and accurate method for the accurate identification of GS and GIST before surgery. |
PDF Full Text Request