基于MRI影像组学机器学习模型在脊髓型颈椎病危险度分级中的价值OA北大核心CSTPCD

Objective:To explore the value of machine learning(ML)model based on MRI radiomics features in grading the risk of cervical spondylotic myelopathy(CSM).Materials and Methods:This retrospective study included 317 patients diagnosed with cervical spondylotic myelopathy(CSM),according to the Japanese Orthopaedic Association(JOA)score they were divided into mild CSM group(193 patients)and moderate-severe CSM group(124 patients).Spinal cord in the transverse T2-weighted MR images were manually sketched to generate a region of interest(ROI)and extract radiomics features.The Z-Score standardization were used to unify metrics.The Pearson correlation coefficients(PCC)and recursive feature elimination(RFE)were used to reduce the dimension and select the feature.Various ML algorithms including logistic regression(LR),adaboost(AB),native bayes(NB),support vector machine(SVM)were used to build ML models.The area under the curve(AUC)of receiver operating characteristic were used to evaluate the diagnostic efficacy of the model.Results:A total of 15 radiomics salient features were selected to build models,SVM(training set AUC vs.test set AUC:0.833 vs.0.813)and LR(0.831 vs.0.812)have good grading ability and are more stable among the four classifiers,there were no statistically significant differences between the models.AB classifier has the best grading ability in the training group(AUC=0.984)but poor grading ability in the test group(AUC=0.725),The AB classifier model has lower stabilization than SVM and LR classifier model.Conclusions:ML model based on MRI radiomics has good risk grading ability for CSM,which can provide certain reference value for clinical preoperative diagnosis.