The Study Of Risk-stratification For Primary Gastrointestinal Stromal Tumors Before And After Surgery

Gastrointestinal stromal tumors(GISTs)are the most common mesenchymal tumors in the gastrointestinal tract.GISTs are clinically heterogenous exhibiting varying degrees of malignant risk in individual patients.The difficulty in assessing the probability of tumors' malignant potential presents challenges to surgeonsTo evaluate their biological behaviors acurrately is significant for the surgeons'decision making:On one hand,accurate assessment before the surgery can help doctor make timely intervention for the progression of tumor,consequentfly increase the chance for cure.On the other hand,accurate assessment after the surgery can help doctor make the optimal decision for the patients who need targeted therapy.Our study aims to evluate the clinical value of the exsiting approaches for risk-stratification and to explore a new method or factor to improve the accuracy in risk prediction before and after sugery separately,providing more effective evidences for precise and individualized treatment of patientsPart ? Preoperative Evaluation of Risk Stratification for the Patients with Primary Gastrointestinal Stromal Tumors1.The Diagnostic Value of CT and EUS in Differentiating Malignant Potential of Gastrointestinal Stromal Tumors PreoperativelyObjective:To investigate the diagnostic value of CT and EUS in differentiating malignant potential of GISTs.Methods:In this study,67 patients with primary GISTs were recruited retrospectively received treatment in our Hospital.CT images and EUS data were both gathered.Stepwise regression model was used to analyze the independent risk factors of GISTs.The ROC curve was used to analyze the CT and EUS signs to judge the malignant potential of GISTs.Results:The AUC of CT in differentiating malignant potential of GISTs was 0.774 slightly higher than that of EUS(0.713).Conclusions:Compared to EUS,CT may be more effective in preoperative assessment of GISTs' risk-stratification.2.Preoperative Prediction Nomogram of Malignant Potential for Gastrointestinal Stromal Tumors based on Radiomics approachObjective:To develop and validate a radiomics nomogram for preoperative predicting the malignant risk of GISTs.Methods:In this study,130 patients with clinicopathologically diagnosed GISTs were recruited retrospectively in our center as a primary cohort.Arterial phase preoperative CT images were adopted for feature extraction.A support vector machine(SVM)classifier was used to select the most important feature subset with good distinguishing characteristics and establish a classification model for malignant risk.Ninty-two patients with GISTs from other hospitals were enrolled as external validation cohort.Clinical and sujective CT models developed in a multivariable logistic regression analysis were compared with our radiomics model and the integrated nomogram.Results:A total of 10320 features were extracted from the tumor volume.Ten optimal features were selected and used to develop the prediction model with an AUC value of 0.867 in the primary cohort and 0.847 in the external cohort.In the entire cohort,the AUCs for the clinical index model,subjective CT findings model,radiomics model,and radiomics nomogram were 0.759,0.774,0.858 and 0.865,respectively.Conclusions:The proposed radiomics model may become an applicable adjunct for predicting the malignant risk of GISTs before operation.Part ? Postoperative Evaluation of Risk Stratification for the Patients with Primary Gastrointestinal Stromal Tumors1.Performance of Different Risk Stratification Systems for Gastrointestinal Stromal TumorsObjective:To evaluate the application value of four different risk stratification systems for the patients with GISTs.Methods:In this study,1.303 Patients who were diagnosed with GISTs and underwent surgical resection without adjuvant therapy in four hospitals were identified from the database.Risk of recurrence was stratified by the modified NIH criteria,the AFIP criteria,the MSKCC prognostic nomogram and contour maps.ROC curves were established to compare the four abovementioned risk stratification systems based on AUCs.Results:According to the ROC curve,the AFIP criteria showed a larger AUC than the modified NIH criteria,the MSKCC nomogram and contour maps criteria.Conclusions:Our data show that the AFIP criteria is likely to be more efficient.2.Application Value of New Nomogram for Risk Stratification in Patients with Primary Gastrointestinal Stromal Tumors based on Ki-67Objective:In our previous study,Ki-67 is a promising predictor of outcome in GISTs.The aim of this study was to build and validate a new risk stratification nomogram for GISTs covering a popular factor Ki-67.Methods:The clinicopathological data of 183 patients with primary GISTs admitted in four hospitals were collected retrospectively.The Log-rank test was performed for univariate analysis of clinicopathologic factors,and COX Proportional Hazard Model was used in multivariate analysis and the nomogram building.Eighty-seven patients from other hospital were used for validation.The application value of this new nomogram was evaluated by comparing with those currently used risk classification methods:modified NIH,AFIP.MSKCC nomogram,and contour maps.Results:In the validation cohort,the AUCs for our nomogram,modified NIH,AFIP,MSKCC nomogram,and contour maps were 0.818,0.758,0.767,0.766,and 0.804,respectively.Conclusions:Our new nomogram might become a potential method to supplement the currently used risk classification criteria for the patients with GISTs.3.Exploring the Application Value of Deep Learning in the Risk Stratification for Primary Gastrointestinal Stromal TumorsObjective:The aim of this study was to explore the role of deep learning in the risk assessment for GISTs.Methods:The clinicopathological data of 41 patients with primary GISTs admitted in four hospitals were collected retrospectively.The simple architecture of ResNet was adopted.The patches of tumor VOI based on the CT from 31 patients were input into our ResNe.Leave-One-Out method was used for validation in training group.Ten patients with GISTs were enrolled for independent validation.Results:The AUC of validation cohort was 0.884.The accuracy of the recurrence risk prediction was 80%.Conclustions:This exploratory experiment shows the feasibility of deep learning in the GISTs risk prediction to some extent,and it also provides reference information for further research in the future.