Preliminary Study On Prognosis Of Acute Cerebral Infa Rction Based On DWI Imaging Radiomics Model

Part 1 Model study of acute cerebral infarction assessment based on DWI radiomics methodObjective: Acute cerebral infarction lesions are heterogeneous and affect the patient's treatment and prognosis.Conventional imaging methods are insufficient to assess this heterogeneity.This study intends to establish an imaging omics model to explore the variability of cerebral infarction from the perspective of big data analysis and determine its possibility to assess the prognosis of cerebral infarction.Methods:The clinical and DWI data of 220 patients with acute middle cerebral artery infarction were collected continuously.The ITK software and support vector machine,Logistic regression analysis and RF random forest classifier model were used to analyze and obtain the best imaging ensemble model.Results:In the final 208 cases,only the ages of the two prognostic clinical and demographic factors were statistically different,and the ages were normally distributed,with an average of 59.053.A total of 396 histological features were extracted between the prognosis group and the poor group.Eight of the strongest correlation features were selected after lasso regression and the best RF-based prediction model was obtained.The area under the ROC curve AUC was 0.824,the sensitivity was 0.953,the specificity was 0.592,the accuracy was 0.699,and the AUC of the verification group was 0.725.The sensitivity,specificity and accuracy were 0.778 and 0.636,0.677,respectively.Conclusion:The features extracted by radiomics method can be used to establish an assessment model of acute cerebral infarction in MCA region based on DWI sequence,and the RF algorithm model is best.Part 2 Preliminary study on prognosis of acute cerebral infarction based on DWI radiomics characteristicsObjective:The purpose of this study was to establish a comprehensive prognostic model for imaging omics and to evaluate the diagnostic efficacy of the model for a good prognosis of acute cerebral infarction.Methods:A total of 208 patients with acute middle cerebral artery(MCA)infarction were enrolled.A total of 396 imaging histological features were extracted.The prognosis was divided into a good prognosis group and a poor prognosis group.Logistic regression and RF random decision forests were used.The classifier model was obtained for each model's receiver operating curve under the ROC area AUC to compare the diagnostic efficiencies of the two integrated models.Finally cross-validation is performed in the verification group.Results:In the simple imaging group,the AUC was 0.82,the sensitivity was 0.953,the specificity was 0.592,and the accuracy was 0.699.The AUC was 0.725,the sensitivity was 0.778,the specificity was 0.636,and the accuracy was 0.677.The imaging group model combined with age factors combined with the predictive model training group AUC was 0.88,sensitivity 0.698,specificity 0.883,accuracy 0.829,and internal cross-validation in the validation group,the AUC of the validation group was 0.87,sensitivity 0.944,specificity 0.705,accuracy 0.774.Conclusion:This study developed a DWI-based imaging prognostic assessment method for acute cerebral infarction,in which the comprehensive model of comprehensive imaging pedagogy and clinical age characteristics predicted the best.