多模态MRI影像组学模型预测非小细胞肺癌脑转移瘤EGFR突变状态的研究OA

Objective:To determine the predictive value of multimodal MRI radiomics models in assessing epidermal growth factor receptor(EGFR)mutation status in non-small cell lung cancer(NSCLC)patients with brain metastases,ultimately providing imaging evidence to guide personalized targeted therapy.Methods:A total of 97 patients with brain metastases were retrospectively included and randomly divided into a training group(40 with mutations and 38 with wild-type)and test group(10 with mutations and 9 with wild-type).The imaging and related clinical data from the pre-treatment MRI scans,including the plain and enhanced scans,were collected using the medical record system.The radiomics features were extracted from the T1WI enhancement,apparent diffusion coefficient(ADC)and T2-fluid attenuated inversion recovery sequence(T2-FLAIR)images.Feature reduction and selection were performed using variance selection and univariate selection methods.Refined features were utilized to build three classifiers-support vector machine(SVM),random forest(RF)and logistic regression(LR)to establish radiomic prediction models.The predictive efficacy of these models for EGFR mutation status was evaluated using receiver op-erating characteristic(ROC)curves.Results:The LR classifier exhibited the best predictive performance in the combined se-quences of T1WI enhancement,ADC,and T2-FLAIR images.The training group showed an area under the curve(AUC)of 0.854,with sensitivity,specificity,and accuracy of 78%,74%,and 69%.In contrast,the testing group achieved an AUC of 0.833,a sensitivity of 83%,a specificity of 71%,and an accuracy of 71%.Moreover,the LR analysis confirmed that both the T1WI enhancement and T2-FLAIR radiomics models showed effective predictive performance,whereas the ADC model yielded less satisfactory results.Conclusions:The radiomics models incorporating the T1WI enhancement and T2-FLAIR imaging can be used to predict the mutation status of EGFR,with the LR combined classifier to exhibit the best predictive performance in the test set(AUC=0.833).This model offers a non-invasive tool for predicting EGFR mutation status in patients with brain metastases who have limited tissue for detection.