Computed Tomography-based Radiomics Model For Predicting Deletion Of KIT Exon 11 Codons 557-558 In Gastrointestinal Stromal Tumors

Objective:To evaluate radiomics model as a tool to differentiate gastro intestinal stromal tumors（GISTs）with deletion of KIT exon 11 codons557-558.Method:A total of 126 GIST patients confirmed by postoperative path ology,who did genetic test were enrolled and their clinical information was collected and analyzed.We extracted and selected radiomics features from three-dimensional volume of interest（VOI）of intravenous CT in GIST patients,and then the radiomics LR model and the fusion model combined radiomics and clinical features were constructed.Receiver operating characteristic（ROC）curve analysis was performed using five-fold cross-validation to evaluate the prediction efficiency of radiomics model and fusion model combined radiomics and clinical features.Results:The non-gastric location,higher mitotic and higher Ki-67 expression were associated with the GISTs with deletion of KIT exon 11 codons 557-558 in the training cohort.The radiomics model,consisting of 12 radiomic features had AUC of 0.9±0.01,0.76±0.04 and 0.85 in the prediction of deletion of KIT exon 11 codons 557-558 in the training,validation and test cohort,respectively.Fusion model combined radiomics and clinical feature had higher prediction efficiency than the former with AUC of 0.93±0.01 in the training cohort,0.82±0.04 in the validation cohort and 0.875 in the test cohort,but there is no statistical difference.Conclusion:The radiomics model can potentially help to predict GISTs with deletion of KIT exon 11 codons 557-558,for guiding treatment selection and prognosis evaluation better.