Establishment And Validation Of Prediction Of Ki67expression In High-grade Glioma Based On MRI Radiomics

Objective: This study aimed to investigate the establishment of radiomics signatures related to Ki67 expression in High-Grade Gliomas(HGG)based on preoperative Magnetic Resonance Imaging(MRI)T1 Weighted Imaging(T1WI)and T2 Weighted Imaging(T2WI).The ultimate goal was to predict and validate Ki67 expression in patients with high-grade gliomas.Methods: In this retrospective study,63 cases of patients pathologically diagnosed with high-grade gliomas from January 2014 to June 2020 were collected.The expression of Ki67 was divided into high and low groups with a threshold of 25%.Preoperative MRI data,including T1 WI and T2 WI sequences of the 63 patients,were collected.Images were imported into the 3D-slicer software in DICOM format.Two experienced radiologists with over 10 years of experience manually delineated the gross tumor volume(GTV)on T1WI/T2 WI images,defined as the region of interest(ROI).If there was a ≥5% discrepancy between the ROIs delineated by the two radiologists,they engaged in further discussions to determine the final tumor boundary.Radiomics features of the delineated GTV were extracted using the Py Radiomics version 2.1 toolkit.Radiomics features with significant differences were selected through t-tests and the least absolute shrinkage and selection operator(LASSO)based on the high and low Ki67 expression groups.The most stable model among seven machine learning models was selected to generate the radiomics signatures Rad＿Score associated with high and low Ki67 expression.A nomogram model was built combining clinical features data and Rad＿Score to examine the relationship between various indicators and Ki67 expression.Results: After applying the inclusion and exclusion criteria,63 patients were finally enrolled,randomly divided into a training set(n=51)and a validation set(n=12).The low expression group comprised 37 cases,and the high expression group comprised 26 cases.A total of 1561 original radiomics features were extracted from both groups of images.After LASSO dimension reduction,12 stable radiomics features were obtained from the T1 WI images and 8 stable features from the T2 WI images.The radiomics signatures Rad＿Score＿T1and Rad＿Score＿T2 were generated using Logistic and SVM models,respectively.The nomogram,incorporating radiomics signatures and clinical indicators,showed a strong correlation between the established radiomics signatures and Ki67 expression,while age,gender,pathological grading levels(III,IV),and EGFR mutation status showed no correlation.The concordance index in the training and validation sets were 0.946(95% CI: 0.887-1)and0.971(95% CI: 0.892-1),respectively.Conclusion: Radiomics signatures associated with Ki67 expression in high-grade gliomas can be effectively established based on preoperative T1WI/T2 WI MRI images.These signatures can be utilized to predict Ki67 expression status in patients with high-grade gliomas preoperatively.