130 Device-Specific Clot Radiomics as Predictors of Revascularization Outcomes for Personalized Thrombectomy in Patients With Acute Ischemic Stroke

View at publisher
Authors: Tatsat Patel, Vinay Jaikumar, Adnan H. Siddiqui, Vincent Tutino
Cited by: 1

Abstract

INTRODUCTION: Radiomic analysis of clot composition from pre-treatment CT imaging offers a non-invasive window into the biological and mechanical properties that influence thrombectomy success. While prior studies have linked clot features to outcomes, the role of treatment device type in modulating these associations remains underexplored. METHODS: We retrospectively analyzed stroke patients who underwent thrombectomy with stent retriever, aspiration, or combination devices at the Gates Vascular institute (n=111). CTA and NCCT were co-registered using BRAINSFit, and clots were segmented using contrast cutoff on CTA and hyperdensity on NCCT. Radiomic features (n=293) were extracted using PyRadiomics from NCCT (bin width = 5), CTA (bin width = 10), and subtracted CTA–NCCT images (bin width = 10). Treatment success was defined as first-pass effect (TICI ≥2b) with the initial device. Top five features were selected using mutual information (univariate) and random forest importance (multivariate) within each device type. UMAP was used to assess feature-based separability, and 5-fold cross-validated AUCs evaluated predictive performance. RESULTS: Of 111 patients, 19 received stent retriever (5 FPE), 30 aspiration (9 FPE), and 62 combination therapy (26 FPE). Univariate analysis revealed minimal overlap across devices, with only one shared feature between stent retriever and combination. Multivariate features were entirely device-specific. Despite lower power, stent retriever yielded an AUC of 0.64, while aspiration and combination groups achieved stronger performance (AUCs: 0.80 and 0.75, respectively). UMAP visualizations showed strong outcome-based clustering for aspiration and combination cohorts. CONCLUSIONS: These results demonstrate that clot radiomics not only predict outcomes but capture distinct device-specific patterns, supporting their use in developing future tools for personalized thrombectomy planning.

Fields of Study & Subjects

Acute Ischemic Stroke Management Radiomics and Machine Learning in Medical Imaging Cerebrovascular and Carotid Artery Diseases

Publication details

Journal / source
Neurosurgery
Publisher
Ovid Technologies (Wolters Kluwer Health)
Volume
72
Issue
Supplement_1
Pages
19–19
Publication date
26 March 2026
Language
EN