Artificial Intelligence Study Of Intracranial Atherosclerotic Stenosis On High-Resolution MRI

Part 1 Identification of Symptomatic Intracranial Plaques with 3D High-resolution Magnetic Resonance Imaging-based Radiomics and Machine LearningBackgroundIntracranial atherosclerotic stenosis is the main cause of ischemic stroke,and the recognition of plaque status is of great significance for stroke prevention and treatment.The aim of this study was to develop a machine learning model for identifying symptomatic intracranial artery plaques using radiomics features based on three dimensional(3D)high-resolution magnetic resonance imaging(HR-MRI),and compare it with traditional model.MethodsA total of 136 patients with intracranial atherosclerotic stenosis admitted for ischemic stroke were retrospectively included.All patients underwent routine MRI and 3D HR-MRI during admission.Clinically and radiographically confirmed culprit lesions of a cerebrovascular event were defined as symptomatic plaques and others outside the culprit vessel as asymptomatic plaques.Radiological plaque features obtained by univariate and multivariate analysis between these two groups were used to construct the traditional model.Radiomics features of plaques were extracted from T1-weighted images(T1WI)and contrast-enhanced T1WI(CE-T1WI).Linear correlation threshold and F-test were used for feature selection.Combined with linear support vector classification(Linear SVC),a machine learning method,the radiomics model was constructed.The radiological and radiomics features were combined to build a combined model.Five-fold cross validation was used to evaluate the generalization ability of the model.Receiver operating characteristic(ROC)and Delong test was used to evaluate and compare the diagnostic performance of the models respectively.ResultsPlaque length(OR=1.122,95%CI:1.013-1.244,P=0.028),burden(OR=1.115,95%CI:1.052-1.182,P<0.001),and enhancement degree(OR=4.475,95%CI:2.123-9.432,P<0.001)were independently associated with clinical symptoms and were included in the traditional model,which had an area under the curve(AUC)of 0.853 vs.0.837 in the training and test sets.While the radiomics and the combined model showed an improved AUC:0.923 vs.0.925 for the training sets and 0.906 vs.0.903 in the test sets.Both the radiomics model(P=0.024,P=0.018)and combined model(P=0.042,P=0.049)outperformed the traditional model in the two sets,whereas the performance of the combined model was not significantly different from that of the radiomics model in the two sets(P=0.583 and P=0.606).ConclusionsThe radiomics model based on 3D HR-MRI can accurately differentiate symptomatic from asymptomatic intracranial arterial plaques and significantly outperforms the traditional model.Part 2 Prediction of Mixed Ischemic Stroke Mechanism Based on 3D High-resolution Magnetic Resonance Imaging Radiomics of Intracranial Arterial PlaqueBackgroundThe vulnerability of intracranial arterial plaque is related to the occurrence of stroke,while the severity of str oke is related to the stroke mechanism.Different mechanisms correspond to different treatment responses and prognosis.Therefore,the understanding of the mechanism of plaques and the stroke relation will help to improve the treatment and prognosis.There are multiple stroke mechanisms in intracranial atherosclerotic stenosis,among which mixed ischemic stroke mechanism has the highest risk of stroke recurrence.This part of the study aims to explore the imaging characteristics of responsible plaques in patients with mixed ischemic stroke mechanism,and establish and verify the radiomics-based model to predict mixed stroke mechanism.MethodsThe three-dimensional(3D)high-resolution magnetic resonance imaging(HR-MRI)and diffusion weight imaging(DWI)data of 137 patients with acute/subacute intracranial atherosclerotic ischemic stroke from November 2016 to January 2022 were retrospectively analyzed.According to the lesion distribution pattern on DWI,the patients were divided into mixed mechanism group and non-mixed mechanism group.Univariate and multivariate analysis were used to analyze the imaging characteristics of responsible plaques in these two groups,and the traditional prediction model was constructed using logistic regression model.The radiomics features of intracranial plaques were extracted based on 3D HR-MRI sequences,and were divided into training set(n=95)and test set(n=42)with a ratio of 7:3 by random sampling.Linear correlation threshold and ANOVA were used for feature selection.The selected radiomics features were used to build a machine learning model.A combined model was built using both the traditional and radiomics features.Receiver operating characteristic(ROC)curve was used to evaluate the diagnostic performance of the model.Delong test was used to compare the prediction performance of each model.ResultsMultivariate logistic analysis showed that the enhancement ratio was an independent predictor of mixed infarction mechanism(OR=2.77,P=0.002).The area under the curve(AUC)of the training set and the test set were 0.676 and 0.568,respectively.The machine learning model composed of radiomics features showed good discrimination ability,with an AUC of 0.906(95%CI:0.849-0.964)in the training set and 0.828(95%CI:0.704-0.951)in the test set.The prediction performance of the combined model was the best,with the AUC of 0.917(95%CI:0.864-0.969)and 0.837(95%CI:0.708-0.966)in the training and test set,respectively.ConclusionThe radiomics model of intracranial arterial plaque based on 3D HR-MRI can effectively predict the mixed stroke mechanism,which is helpful to take targeted clinical treatment measures.