Predicting The Prognosis Of Acute Basilar Artery Occlusion Treated With Endovascular Treatment Based On Multimodal Medical Imaging

Acute basilar artery occlusion(ABAO)is a fatal stroke subtype with high mortality and morbidity.However,the proportion of favorable prognosis in ABAO was significantly lower than that of anterior circulation stroke,even though they received the identical procedures of endovascular treatment.The remarkably differential anatomical and pathophysiological features of posterior circulation stroke from anterior circulation stroke are insufficient to explain the phenomena of poor clinical outcomes of ABAO treated by endovascular treatment.Based on previous researches,continuous optimization of clinical-decision making systems for endovascular treatment in anterior circulation stroke has significantly improved their long-term prognosis,which implied that the lack of accurate prognosis prediction models for ABAO treated with endovascular treatment might act essential roles in the poor clinical outcome of these patients.Previous literature also showed that it was not appropriate to apply current preoperative outcome evaluation systems of anterior circulation stroke in ABAO.Given the lack of prognosis prediction system for ABAO patients after endovascular treatment,this study intends to explore novel predictors for ABAO treated by endovascular treatment,and try to establish a mortality prediction model for these patients based on multimodal images and machine learning algorithms.Part 1 Effects of brain atrophy on outcomes of acute basilar artery occlusion patients receiving endovascular treatment based on CT images and automatic imaging segmentation algorithmBackground and purpose: The current preoperative evaluation system for ABAO patients mainly relied on the PC-ASPECTS score derived from Computed tomography(CT)images,which could provide rapid but initial assessment of damage degrees according to the localization of these infarcts.In addition to the degrees of injury,the reconstruction and recovery ability of brain tissue after stroke is of great value in determining the clinical outcome of patients.Brain atrophy globally reflects the effects of preexisting risk factors and biological aging on brain structures and could predict poor outcomes in anterior circulation stroke.However,comparing with these patients,acute basilar artery occlusion(ABAO)impairs infratentorial regions frequently and might benefit from brain atrophy due to the resulting residual space to reduce tissue compression,which raises doubts that current understandings for prognostic roles of brain atrophy are also applicable for ABAO.Therefore,this study aims to evaluate brain atrophy automatically from CT images and investigates its impacts on outcomes of ABAO following endovascular treatment.Methods: A total of 231 ABAO who underwent EVT in our preceding registry study were enrolled in present research.Brain atrophy was quantified as the ratio of brain parenchymal volume to cerebrospinal fluid volume on CT slices at admission.The primary outcome was determined by the modified Rankin Scale(m RS)score at 3 months.Results: The frequency of favorable outcomes(90-day m RS≤3)was significantly decreaed in the severe atrophy group(P=0.014).Restricted cube analysis showed that there existed a significant nonlinear correlation between the degrees of brain atrophy and the probability of favorable outcome.The distribution of confidence interval of severe brain atrophy group was significantly different from those with non-severe brain atrophy(P=0.034).Adjusted logistic regression analysis showed that severe brain atrophy was significantly negatively correlated with the incidences of good outcomes(P=0.002),but not with mortality at hospitalization or at 90 days(P>0.05).Adding a severe atrophy index into the baseline model obviously enhanced its discriminatory ability in predicting the outcome by remarkably increasing areas under the receiver operating characteristic curve,net reclassification improvement algorithm,and integrated discrimination improvement algorithm values(all P<0.05).Conclusion: Severe brain atrophy did not improve in-hospital or overall mortality,but impaired the long-term recovery after EVT.This objective and automated marker has the potential to be incorporated into decision-support methods for ABAO.Part 2 Predicting of prognosis of acute basilar artery occlusion with endovascular treatment based on preoperative multimodal magnetic resonance imaging featuresBackground and purpose: Comparing with the impaired regions in anterior circulation stroke,the brain structures supplied by basilar artery are brainstem,thalamus and cerebellum.As important parts of conduction tracts and neural circuits,these structures acted key roles in maintaining a variety of neurological functions,such as consciousness,movement,and language,resulting in significantly differential clinical manifestations and outcomes in patients with different degrees of injury in the same brain area.The current preoperative evaluation system for ABAO patients relied on the PC-ASPECTS score derived from computed tomography(CT)images,which provided location information of infarscts,but neglected the detailed description of impaired degrees in specific regions.In addition,the bone of the posterior cranial fossa is uneven,which would lead to bone artifacts and limited the ability of CT to analyze the quantitative information of infarcts.Comparing with CT,magnetic resonance imaging(MRI)imaging technology has a higher soft tissue resolution ability than CT,which can effectively overcome the problems of posterior fossa imaging,and is one of the most useful techniques to quantify infarcts volume.Therefore,considering the important roles of infarct localization and volume quantitative information in determining the clinical outcome of stroke patients,this study aimed to explore the prognostic neuroimaging predictors for ABAO patients receiving endovascular treatment based on preoperative multimodal magnetic resonance imaging.Methods: 54 ABAO patients with preoperative multimodal magnetic resonance imaging in the multicenter real-world study were included in this study.The volumes of cortical and subcortical brain regions,the location and volume quantitation information of infarcts and the volumes of white matter hyperintensity(WMH)were obtained by full-automatic and semiautomatic segmentation methods.The prognostic values for these structures were determined by ANCOVA analysis and logistic regression analysis.The main clinical outcome was determined by the modified Rankin Scale(m RS)score at 3 months.Results: Based on automatic or semi-automatic segmentation methods,this study obtained the volumes of brain regions and the infarct volumes distributed in each region of PCASPECTS.Among these factors,the decreased ratio of infarct volume in midbrain and pons to total brainstem volume(RMPI)was significantly correlated with the favorable outcome.Based on logistic regression statistics,NIHSS score and RMPI were independent predictors for favorable outcome.Conclusions: Through multimodal MRI,we found that the ratio of infarcts volume in midbrain and pons to total brain stem volume might be a novel predictor for favorable outcome among ABAO after endovascular treatment.Part 3 Machine learn algorithms-based mortality prediction model for acute basilar artery occlusion after endovascular treatmentBackground and purpose: The 90-day overall mortality of patients with acute basilar artery occlusion(ABAO)who received endovascular treatment was 39.6%.The death patients accounted for nearly 70% of ABAO individuals with poor prognosis,while the mortality of patients with anterior circulation stroke was less than 20% and the death population of them only accounted for about 40% of anterior circulation stroke patients with poor prognosis.The high mortality rate of ABAO patients and the high proportion of dead patients in the population with poor prognosis made the mortality prediction to be essential for clinical decision-making among these individuals.Reducing mortality of patients also acted important roles in further improving the prognosis of ABAO patients receiving endovascular treatment.However,none present studies focused on the prediction of mortality of ABAO patients after endovascular treatment.Methods: Patients in southern China were enrolled in training cohort,while the others were enrolled in testing group based on the significant differences in geographical environment and living habits between these two regions.We compared the prediction ability of nine machine learning algorithms in the southern patient cohort.Among all algorithms,xgboost algorithm achieved the best prediction performance,which was then applied to construct the mortality prediction model.The performance of the final prediction model was verified in the northern patients cohort.Results: In both of the training cohort and the validation cohort,the prediction performance of POSITIVE model is significantly better than those of other traditional prediction models.According to decision curve analysis,the benefits of patients in both of the training cohort and validation cohort in the POSITIVE model were significantly better than in the traditional models.Based on density distribution maps,death and survival populations were widely distributed in the POSITIVE model.Further subgroup analysis found that the mortality rate was significantly higher in the high-risk group than that in the low-risk group in the POSITIVE model(P<0.05).Conclusion: Through the model construction,model validation and comparisons with previous models in the multicenter cohorts,the prediction performance of POSITIVE prediction model was much better than previous algorithms,which might participate in decision-making process of ABAO.Part 4 Incidence and predictors for malignant cerebellar edema and its impacts on the prognosis of acute basilar artery occlusion after endovascular treatment based on CT imagesBackground and purpose: The prognosis of ABAO patients was still poor even if they received endovascular treatment,which suggested the necessities of further exploring the factors limiting the efficacy of endovascular treatment.Through the analysis of postoperative images in the ABAO database previously established by our group,we found that the cerebellar infarcts in ABAO patients may progress to malignant cerebellar edema,causing secondary damage to the brainstem through space occupying effects.Therefore,the purpose of this study was to investigate the effects of malignant cerebellar edema on the long-term prognosis of ABAO patients,and to explore the predictors of malignant cerebellar edema in ABAO patients after endovascular treatment.Methods: A total of 329 ABO patients with cerebellar infarctions treated by EVT met inclusion criteria.The presence of malignant cerebellar edema defined by Jauss scale≥4 points,was evaluated on the computed tomography performed 72 hours after EVT.Adjusted odds ratio and 95% CI were obtained by logistic regression models.A favorable outcome was defined as a 90-day modified Rankin Scale score of 0–3.Results: Malignant cerebellar edema was statistically associated with decreased incidence of favorable outcome(adjusted odds ratio,0.35[95% CI,0.18-0.68],P=0.002).Baseline NIH Stroke Scale score,collateral circulation,neutrophil count at admission and recanalization status were predictors for malignant cerebellar edema and favorable functional status at 90 days(all P<0.05).Among all inflammatory factors,neutrophil counts achieved the highest accuracy,sensitivity,and specificity for malignant cerebellar edema(AUC,0.68[0.62-0.74]).Adding neutrophil count status into the baseline model obviously enhanced its prediction ability for malignant cerebellar edema and favorable outcome by increasing area under curve,and achieving both net reclassification and integrated discrimination improvement(all P<0.05).Mediation analysis indicated that malignant cerebellar edema mediated the association between increased neutrophil count and worse functional outcome(P=0.026).In the further sensitivity analysis,after excluding patients with symptomatic hemorrhage,any bleeding events,and failure to obtain a good recanalization status,high neutrophil counts were still significantly associated with malignant cerebellar edema and poor prognosis(P<0.05).Conclusions: The occurrence of malignant cerebellar edema played important roles in the deterioration of clinical outcome of ABAO patients.Predictors of malignant cerebellar edema included onset to recanalization time,baseline NIHSS,collateral circulation,neutrophil count at admission and recanalization status.Compared with immutable factors such as the severity of acute cerebral infarction and the state of collateral circulation,neutrophil counts at admission can not only be included in the clinical decision-making system as an easily available biomarker in blood routine examination,but also be used as a target to guide the immunomodulatory treatment of ABAO.