Time-variant and tissue-level collaterals predict postoperative neurological recovery and clinical outcomes of patients with endovascular thrombectomy

Abstract

Background: A comprehensive assessment of collateral status can yield profound insights into the ischemic mechanism in patients experiencing acute ischemic stroke. This study aims to investigate whether time-variant and tissue-level collateral characteristics may serve as predictors for functional outcomes in patients undergoing endovascular thrombectomy (EVT) through the application of machine learning (ML) algorithms, and to stratify postoperative neurological recovery of these patients.

Methods: In this retrospective study, 128 acute ischemic stroke patients characterized by anterior large-vessel occlusion and received EVT between May 2020 and December 2022 were enrolled. These patients underwent multiphase computed tomography (CT) angiography (mCTA) and CT perfusion (CTP). The time-variant collateral score was defined as the Collateral Score on Color-Coded summation maps (CSCC) of mCTA. The hypoperfusion intensity ratio (HIR) was calculated from CTP data. The data were split into training and test sets in a ratio of 7:3, and univariable and multivariable regression analyses were employed for feature selection. For ML analyses, logistic regression (LR), support vector machine (SVM), random forest (RF), decision tree (DT), and eXtreme gradient boosting (XGBoost) algorithms were utilized. The receiver operating characteristic (ROC) curve and decision curve were employed for performance evaluation. The mixed effect model was established to estimate the impact of collateral stratification on the postoperative National Institutes of Health Stroke Scale (NIHSS).

Results: Age [odds ratio (OR) =1.073; 95% confidence interval (CI): 1.008, 1.154; P=0.040], Alberta Stroke Program Early CT Score (ASPECTS) (OR =0.742; 95% CI: 0.546, 0.975; P=0.040), CSCC (OR =0.468; 95% CI: 0.213, 0.953; P=0.044), and HIR (OR =56.666; 95% CI: 3.843, 1,156.959; P=0.005) were significantly associated with good outcome in training set. By utilizing these four selected features, the RF algorithm achieved the best performance and the highest clinical suitability in predicting good clinical outcomes, with an area under the ROC curve (AUC) of 0.964 (95% CI: 0.902, 0.992) and 0.837 (95% CI: 0.684, 0.935) in training set and testing set, respectively. The Shapley Additive exPlanations (SHAP) analysis revealed that HIR was the most significant variable in predicting clinical outcomes. Fixed effects and group × time interaction effects [all P<0.01 at all time points (TPs)] were acquired in HIR stratification. HIR enabled better stratification and prediction of patients' postoperative NIHSS [Akaike information criterion (AIC): HIR =4,599.577 and CSCC =4,648.707].

Conclusions: RF model, which has been trained on time-variant and tissue-level collaterals, is capable of accurately predicting the clinical outcomes of patients undergoing EVT. Stratifying patients based on HIR may yield valuable insights into predicting trends in the potential postoperative neurological recovery.

Keywords: Ischemic stroke; collateral circulation; computed tomography angiography (CTA); machine learning (ML); thrombectomy.

Figure 1

Patient flowchart. CTA, computed tomography angiography; CTP, computed tomography perfusion; EVT, endovascular thrombectomy; ICA, internal carotid artery; mRS, modified Rankin Scale; TOF-MRA, time of flight magnetic resonance angiography.

Figure 2

Examples of imaging in patients with good functional outcome and poor functional outcome. A 56-year-old man with left weakness and lisp for 9 hours who has M1 segment occlusion on right MCA (A). The follow-up 90-day mRS was 1. CSCC on mCTA was Score 3. Hypoperfusion area was seen on Tmax map with HIR of 0.282. Cerebral infarction in right basal ganglia was shown on non-contrast CT on admission. After EVT, the M1 segment is recanalized successfully. A 60-year-old man with right weakness and lisp for 13.5 hours has M1 segment occlusion on left MCA (B). The follow-up 90-day mRS was 4. CSCC on mCTA was Score 2. Hypoperfusion area was seen on Tmax map with HIR of 0.542. Cerebral infarctions in left basal ganglia, left temporal lobe and left insular lobe were shown on non-contrast CT on admission. Successful recanalization was not achieved on M1 occlusion. The arrows indicate the recanalization status on DSA following EVT treatment. CSCC, Collateral Score on Color-Coded summation maps; CT, computed tomography; DSA, digital subtraction angiography; EVT, endovascular thrombectomy; HIR, hypoperfusion intensity ratio; MCA, middle cerebral artery; mCTA, multiphase computed tomography angiography; mRS, modified Rankin Scale; Tmax, time to maximum.

Figure 3

ROC curves and decision curves for ML models for prediction of clinical outcomes in training set (A,B) and testing set (C,D). (A,C) The X-axis represents the false positive rate (100 – specificity), which indicates the proportion of negative cases incorrectly classified as positive. The Y-axis represents the true positive rate (sensitivity), which shows the proportion of actual positive cases correctly identified by the model. AUC, area under the ROC curve; DT, decision tree; LR, logistic regression; ML, machine learning; RF, random forest; ROC, receiver operating characteristic; SVM, support vector machine; XGBoost, eXtreme gradient boosting.

Figure 4

Comparison of ML model performance for the prediction of clinical outcome. Y-axis represents the performance scores (ranging from 0 to 1) for each evaluation metric (AUC, sensitivity, specificity, accuracy, precision, recall, and F1-score) across different ML models. AUC, area under the ROC curve; DT, decision tree; LR, logistic regression; ML, machine learning; RF, random forest; ROC, receiver operating characteristic; SVM, support vector machine; XGBoost, eXtreme gradient boosting.

Figure 5

Confusion matrix and SHAP plot for RF model. ASPECTS, Alberta Stroke Program Early CT Score; CSCC, Collateral Score on Color-Coded summation maps; HIR, hypoperfusion intensity ratio; RF, random forest; SHAP, Shapley Additive exPlanations.

Figure 6

Line charts and three-dimensional simulations of NIHSS measures were established to estimate trend of NIHSS in CSCC stratification and HIR stratification, respectively. *, P<0.05; **, P<0.01. CSCC, Collateral Score on Color-Coded summation maps; HIR, hypoperfusion intensity ratio; NIHSS, National Institutes of Health Stroke Scale; TP, time point.