Does the occlusive sign play a role in endovascular thrombectomy: a radiomics-based approach.

Abstract

Background and purpose: Large vessel occlusion (LVO) is a major cause of acute ischemic stroke (AIS). Identifying its underlying etiology, particularly intracranial atherosclerotic stenosis (ICAS), is crucial for optimizing endovascular thrombectomy (EVT). Intra-procedural occlusive signs can offer clues, but their interpretation is often subjective. This study proposes a radiomics-based approach to objectively characterize angiographic signs and predict occlusion etiology in real time.

Methods: We retrospectively included 465 EVT-treated patients with acute M1-segment MCA occlusion from two centers (January 2018-December 2023). Radiomics features were extracted from angiographic parametric imaging (API) and used to develop a radiomics score via least absolute shrinkage and selection operator (LASSO) logistic regression. The score's predictive value for ICAS-LVO was assessed using logistic regression, and the optimal cut-off was determined via the Youden index. Subgroup analyses were performed to compare procedural outcomes between radiomics-inferred ICAS and embolic occlusions.

Results: The radiomics score was significantly higher in ICAS-related occlusions than in embolic occlusions (median 0.39 vs 0.89, P<0.001) and was the strongest independent predictor of ICAS etiology (adjusted odds ratio (OR) 25.40, 95% CI 12.13 to 56.94, P<0.001). Key discriminative features included texture-based parameters from perfusion maps. Based on the Youden index, a cut-off of 0.569 was defined to stratify cases into radiomics-inferred ICAS and embolic groups. Among patients treated with contact aspiration, those with radiomics-inferred ICAS occlusion had lower first-pass reperfusion rates compared with those with radiomics-inferred embolic occlusion (35.6% vs 60.7%, P-value Bonferroni correction =0.004).

Conclusion: Radiomics features extracted from API offer an objective method for intra-procedural inference of occlusion etiology, particularly ICAS-LVO. This approach may support technical efficacy and procedural planning during EVT, especially in populations or regions with higher ICAS prevalence.