MRI Radiomics In The Differential Diagnosis Of Cerebral Alveolar Echinococcosis,Brain Metastases And Cerebral Tuberculomas

Objective: This paper mainly studies the application of radiomics based on machine learning in the differential diagnosis of patients with cerebral alveolar echinococcosis,brain metastases and brain tuberculomas,so as to improve the accuracy of differential diagnosis of diseases.Methods: We collected 40 cases(203 lesions)of cerebral alveolar echinococcosis,61 cases(115 lesions)of brain metastases and 22 cases(73 lesions)of brain tuberculosis who met the inclusion criteria in our hospital from September 2018 to September 2020.All patients underwent MRI plain scan and contrast-enhanced scan.1226 features were extracted by radiomics,and five important features were selected by using LASSO algorithm and random forest machine learning method.Then six machine supervised learning methods such as LR,LD,KNN,SVM,DT and NB were used to establish the model.Finally,Accuracy,AUC,F1-measure and confusion matrix are used to evaluate the performance of the model.Results: Among the six machine learning methods,the performance of LD method was relatively better.The accuracy of discriminating three kinds of diseases in the training set was0.825,the AUC=0.913,and the F1-measure=0.803.In the test set,the discriminant accuracy= 0.759,the AUC=0.889,F1-measure=0.691.Verification of CAE lesions precision=0.804,recall=0.902,verification of BM lesions precision=0.783,recall=0.739,verification of BT lesions,precision=0.667,recall=0.400.LD and LR methods have relatively high quantitative index values in verification set and test set,and LD method is better than LR method in generalization and robustness.Generally speaking,LD method is better for test set classification,LR method is the second,and SVM method test set is less effective than LR method.The results of confusion matrix show that the prediction accuracy of LD method is higher,which is consistent with the results of quantitative indicators.Conclusion:The radiomics based on machine learning can differential diagnosis the three kinds of diseases,and the model established by LD method has the best performance.This model has high clinical application value.