Risk Stratifications Prediction Of Gastrointestinal Stromal Tumors Based On Radiomics

Gastrointestinal stromal tumors(GIST)are the most common mesenchymal tumors of the gastrointestinal tract with non-directional differentiation and varying malignancy potential and lack specificity.So risk stratifications diagnosis of GIST provides guidance for preoperative treatment and prognosis.However,the diagnosis results of clinical identification techniques such as pathological biopsy or CT images are affected by subjective factors and human factors.As an emerging research method for quantifying tumor heterogeneity,radiomics has been widely used in the field of assisted tumor diagnosis,staging and prognosis prediction etc.in recent years due to its non-invasiveness and reproducibility.Radiomics high-throughput extracts a large number of quantitative image features from medical images and combines with statistics,pattern recognition,machine learning and others to conduct objective modeling analysis.It assists in solving clinical disease classification diagnosis,staging classification,predicting prognosis,treatment planning and treatment response evaluation,etc.and is expected to achieve future personalized precision medical treatment.This paper proposed risk stratifications prediction of GIST based on radiomics to explore the its value in assisting clinical tumor staging.GIST patients(13 cases of extremely low risk,48 cases of low risk,27 cases of intermediate risk and 32 cases of high risk)were retrospectively enrolled in this study.In paper,the two-category prediction model and the four-category prediction model were established for gastrointestinal stromal tumors.In the two-category experiment,extremely low risk and low risk were divided into benign,intermediate risk and high risk were divided into malignant.First,the experiment extracted 4 non-textured features and 323 texture features from the three-dimensional tumor region of each case and the original feature set with redundancy and irrelevance was obtained.It was necessary to perform feature selection to improve model classification performance and generalization ability.Then this set is randomly divided into a training set(70%)and a test set(30%).The training set applied the 10-fold cross-validation to select the optimal feature subset and constructed the classification prediction model.The test set was only used to evaluate and validate the final model.During the processing of the original feature set,ReliefF,MI,Fisher Score,EC and forward selection were executed sequentially for feature selection to obtain the optimal feature subset.Finally,SVM,RF,KNN classifiers for risk stratifications of GIST were trained by the optimal feature subset and verified with the test set.In this paper,four feature selection methods and three classifiers were used to radiomics modeling analysis for risk stratifications of GIST.The experimental results show that the classification performance of KNN classifier trained by feature subsets selected by ReliefF based forward selection algorithm was best in the two-class problem.In the two-class experiment,the AUC,sensitivity,specificity,accuracy,positive predictive value and negative predictive value of the model respectively were 0.8988,0.8200,0.8250,0.8139,0.8583 and 0.8388 on the training set;0.8256,0.8050,0.8350,0.8236,0.7944 and 0.8748 on the test set.In the four-class experiment,the performance of the RF model was slightly better than the performance of the other two models.The accuracy of the model on the training set and the test set respectively were 0.6662 and 0.6389.It show that the model established by the radiomics method provides a noninvasive detection method for predicting the risk stratifications of GIST.And this mothed maybe as an auxiliary diagnosis tool to improve the accuracy efficiently and provides guidance for preoperative treatment and prognosis of GIST.