Predictive Value Of The Radiomics-Based Hyperdense Middle Cerebral Artery Sign For The Prognoses Of Patients With Emergency Thrombectomy

Objective: We aim to assess the performance of radiomics-based HMCAS for predicting the long-term prognoses of patients with acute ischemic stroke after MT in the emergency department,and search for other related clinical features before and after MT to provide a reference for selecting patients who can benefit more from MT treatment by predictive model established.Methods: We performed a retrospective study of 102 consecutive patients presented with HMCAS on initial Non-contrast Computed Tomography（NCCT）that had undergone MT in the Emergency Department of the First Hospital of Jilin University between January 2019 and December 2020.Patients were divided into two groups based on the 3-month follow-up after operation,defined as a good prognosis group（modified Rankin scale,m RS≤2）and a poor prognosis group（m RS>2）.46 patients achieved good prognoses（m RS≤2）.Then the patients were divided into the training set（n=81）and test set（n=21）by 8:2.The quantitative radiomic features（RFs）were selected by the analysis of variance（ANOVA）and least absolute shrinkage and selection operator（LASSO）algorithm in the training set.Clinical features（CFs）data before and after MT were analyzed.Statistical analyses were performed to identify variables associated with the prognoses of MT in the training cohort from which 5 Logistic regression machine learning models were established to predict the adverse outcomes before and after MT respectively,and then the nomograms were created.Finally,the predictive abilities of the 5 models were quantified using the area under the receiver operating characteristic curve（AUC）and confirmed via decision curve analysis（DCA）.Results:The clinical prediction model based on the pre-MT clinical variables including the admission NIHSS,hemoglobin,NLR,D-Dimer and other clinical features was evaluated（AUC（95% CI）= 0.596（0.312-0.881）in the test dataset）.The 1389 RFs were extracted from each HMCAS territory,and finally 8 RFs were left to build the radiomic model（AUC（95% CI）=0.798（0.598-0.998）in the test dataset）.The pre-MT model combined with CFs and RFs further achieved good performance（AUC（95% CI）= 0.817（0.625-1.000））in the test dataset.As comparison,two post-MT models including CFs alone,or combining RFs and CFs were established,which showed good discrimination（AUC（95% CI）= 0.856（0.685-1.000）and 0.894（0.754-1.000）in the test dataset respectively）.Conclusions: 1.Radiomic features of HMCAS have limited ability to predict poor prognoses for patients with Emergency Thrombectomy,so some clinical features should also be considered.2.Postoperative prediction performs better,while preoperative prediction also has certain reference value for identifying patients who are most likely to benefit from MT.