Preprocedural Curvature Assessment Can Stratify Recanalization Risk Before Device Selection in Mechanical Thrombectomy
Related Experiment Videos
Jin Vivian Lee1,2,3, Anna L Huguenard1, Ralph G Dacey1
1Department of Neurological Surgery, Washington University School of Medicine, St. Louis, MO (J.V.L., A.L.H., R.G.D., J.W.O.).
Background:
Vascular geometry may impede catheter navigation during mechanical thrombectomy. We evaluated whether cervical curvature predicts recanalization success.
Methods:
This is a retrospective single-center cohort of 160 anterior large vessel occlusion patients treated with mechanical thrombectomy at a comprehensive stroke center (Barnes-Jewish Hospital/Washington University School of Medicine, St. Louis, MO) from May 2019 to February 2021. Cervical curvature was computed using 3-dimensional vessel centerlines derived from preprocedural computed tomography angiography. Primary outcomes were successful recanalization (final modified Thrombolysis in Cerebral Infarction score ≥2b) and intracranial access time (puncture→first pass). Associations were evaluated with multivariable linear or logistic regression, adjusting for age, National Institutes of Health Stroke Scale, first-line device, and access route. Random Forest and Gradient Boosting models with Gini Importance and Shapley Additive Explanation explained feature importance.
Results:
Mean age was 67±15 years; median National Institutes of Health Stroke Scale 16; mean total curvature was 9.80±2.85. Median intracranial access time was 24 minutes (interquartile range, 17-32), with no correlation to cervical curvature (ρ=0.054; adjusted β=0.20 min/unit [95% CI, -0.75, 1.15]; P=0.676). Successful recanalization occurred in 86% and varied across curvature tertiles (low 91%, mid 93%, high 75%; P=0.017). Per-unit curvature increase was associated with recanalization failure (adjusted odds ratio, 1.21 [95% CI, 1.02-1.43]; P=0.028). Receiver operating characteristic-derived curvature threshold (>11.54) identified a high-risk group (adjusted odds ratio, 5.10 [95% CI, 1.88-13.84]; P=0.001). Machine learning confirmed cervical curvature as the dominant predictor, with a Shapley Additive Explanation peak effect at 11.65.
Conclusions:
Cervical curvature independently predicts mechanical thrombectomy recanalization failure, with a threshold effect near 11.6 units. Preprocedural curvature assessment derived from routine computed tomography angiography may improve risk stratification and inform mechanical thrombectomy strategy.