Preoperative Assessment Of Ki-67 Expression Level In Gastric Cancer Based On CT Radiomics

Research Objectives:Objective To evaluate the value of preoperative prediction of ki-67 expression level in gastric cancer based on CT radiomics.Research methods:A total of 500 patients with gastric cancer who underwent enhanced abdominal CT scan and underwent postoperative immunohistochemical staining of tumor tissue from January 2017 to October 2021 were collected.After a series of inclusion and exclusion criteria,70 patients were finally selected to be included in the group,with complete imaging and clinical data.The samples were divided into "high expression group"（Ki-67≥ 50%）and "low expression group"（Ki-67<50%）,with 35 cases in each group.By two experienced radiologists use platform（https://keyan.deepwise.com）,select venous phase images and layered paint region of interest of tumor lesions（ROI）,then the feature extraction and dimensionality reduction on the platform,screening and gastric cancer Ki-67high/low expression level of characteristics of radiomics.Finally,five optimal characteristics of radiomics were screened and radiomics labels（RS-V）were established.SPSS25.0 statistical software and R software were used to analyze all the collated clinical data,which were selected and further incorporated into the clinical prediction model.Then the two models are combined to build a joint model and visualized as a Nomogram.Then the area under curve（AUC）was obtained.De Long’s test was used to compare ROC of different models in pairs,and p < 0.05 was considered as statistically significant difference.AUC value was used to evaluate the classification performance of the prediction model,calibration curve and decision curve analysis（DCA）,sensitivity,specificity,positive predictive value and negative predictive value were used to explore the diagnostic and predictive efficiency of the model.Results:Based on the five texture features of the venous stage,we constructed the radiomics label（RS-V）and combined model,which has good predictive value for the expression of KI-67 in gastric cancer patients.The area under receiver operating characteristic curve（ROC）of the RS-V model in the training and verification groups was 0.900（95%CI: 0.829-0.971）and 0.867（95%CI: 0.867）,respectively.0.782-0.952),accuracy were 0.828 and 0.785,specificity were0.800 and 0.800,sensitivity were 0.857 and 0.771,positive predictive values were0.810 and 0.794,negative predictive values were 0.848 and 0.778,respectively.The area under the receiver operating characteristic curve of the combined model in the training group and verification group was 0.920（95%CI:0.851-0.983）and 0.910（95%CI: 0.833-0.977）,sensitivity were 0.828 and 0.771,accuracy were 0.828 and 0.785,specificity were 0.828 and 0.777,positive predictive values were 0.828 and 0.794,negative predictive values were 0.828 and 0.777,respectively.Therefore,the intravenous phase based imaging label and combined model can predict the expression status of KI-67 in gastric cancer before surgery,which is expected to be a non-invasive prediction method.Conclusion:The combined radiomics model based on CT enhanced venous images can effectively predict the expression level of Ki-67 in gastric cancer.It also provides a new noninvasive method for the evaluation of ki-67 proliferation index in gastric cancer cells.