Abstract
BACKGROUND: Vascular geometry plays an important role in stroke thrombectomy outcomes, but few studies have examined complex characteristics of vessel morphology. The authors hypothesized that engineered vessel cross-section features could be used to predict thrombectomy first-pass effect (FPE). METHODS: The authors analyzed computed tomography angiography and noncontrast computed tomagraphy from 50 patients with anterior circulation stroke thrombectomy. After segmentation, traditional metrics (vessel tortuosity and angulation) were calculated from vessel centerlines that were transformed into the same coordinate system and same region of interest. Univariate statistical analysis and geometric morphometrics were used to interrogate differences in geometry between cases that did and did not achieve FPE, which classical angulation and tortuosity did not quantify. To describe these differences, complex cross-section features were engineered and quantified using a semiautomatic pipeline. Machine learning was used to train predictive models of FPE based on significant cross-section features. RESULTS: Only one local tortuosity metric was significantly different (q=0.019) between FPE and first-pass failure cases. The most significant principal component score (q=0.012) from geometric morphometrics highlighted the M1 segment of the middle cerebral artery and upper cavernous internal carotid artery variation as morphological indicators of first-pass outcome. Fifteen cross-section features, corresponding to internal carotid artery and middle cerebral artery regions, were significantly different between cases that did and did not achieve FPE. Predictive models exhibited a strong prediction of FPE (area under the curve=0.98±0.05) and outperformed models using traditional tortuosity and angulation features. CONCLUSION: Cross-section features are a novel class of powerful and interpretable predictors of FPE, which could assist in treatment decision-making.