Application Of MRI Radiomics In Predicting Functional Outcome Of Patients With Acute Ischemic Stroke

Objective : The diagnostic value of extracting radiomics features of ischemic stroke based on pre admission DWI images and establishing predictive models to predict the clinical prognosis of patients after intravenous thrombolysis with rt-PA.Methods : A retrospective analysis was conducted of 98 patients with acute ischemic stroke diagnosed by DWI at the Affiliated Hospital of Beihua University from January 1,2021 to January 1,2023,and treated with rt-PA intravenous thrombolysis in our hospital.Patients were randomly divided into a training set and a test set,with a ratio of 7:3.On the Philips ISD post processing platform,the infarcted area(DWI high signal)of the training set DWI sequence was delineated and all imaging features were extracted.The t test,minimum absolute shrinkage and selection operator algorithm(LASSO)are used to reduce the dimension of features and filter them to extract the best image radiomics features.Four classifier models,namely,Random Forest(RFC),Support Vector Machine(SVM),Logical Regression(LR)and clinical-radiomics,were constructed to predict the receiver operating characteristic curve(ROC),sensitivity,specificity,and accuracy of the training set and the test set,respectively,and compared to select the best predictive model for predicting the prognosis of AIS patients after intravenous thrombolysis with rt-PA.Results:1.There was no statistically significant difference in clinical data between the training set and the test set groups(P>0.05).The single factor logistic regression analysis of clinical information showed that the NIHSS score on admission was an independent factor in judging the prognosis.2.Using the ISD post processing platform,1199 features were extracted from DWI images.After t-testing and the Least Absolute Shrinkage and Selection Operator Algorithm(LASSO),9 optimal image omics features were finally selected,including 4 belonging to the Gray Level Co occurrence Matrix(GLCM),4 belonging to the Gray Level Run Matrix(GLRLM),and 1 belonging to the Gray Level Region Size Matrix(GLSZM).3.The clinical-radiomics prediction model is the best prediction model among the three prediction models,with a receiver operating characteristic curve(ROC)of0.727,and a sensitivity,specificity,and accuracy of 1,0.545,and 0.66,respectively.Conclusion : This study is based on the prediction model constructed by DWI imaging genomics for AIS patients after using rt-PA intravenous thrombolysis treatment,and selects the best algorithm.It proves that the clinical-radiomics prediction model can better predict the clinical prognosis of AIS patients using rt-PA intravenous thrombolysis,providing a basis for clinical follow-up personalized treatment.