Cerebrovascular CTA radiomics for objective collateral grading in acute ischemic stroke

Area of Science:

• Neuroradiology
• Medical Imaging Analysis
• Artificial Intelligence in Medicine

Background:

• Collateral circulation is crucial for functional outcomes in large vessel occlusion (LVO) stroke.
• Current collateral grading methods are subjective and prone to errors.
• Objective collateral scoring is needed to inform thrombectomy decisions.

Purpose of the Study:

• To develop and validate a fully automated cerebrovascular radiomics pipeline for objective collateral scoring on computed tomography angiography (CTA).
• To compare the performance of the radiomics pipeline against existing methods.

Main Methods:

• Retrospective analysis of CTAs from 343 LVO patients (MR CLEAN trial).
• Development of nnU-Net models for arterial tree and Circle of Willis (CoW) segmentation.
• Extraction of radiomics features and training of a random forest classifier for collateral status prediction.
• Validation on an external cohort of 140 acute LVO patients.

Main Results:

• Accurate segmentation of cerebral arteries and CoW segments achieved.
• The vessel-tree radiomics model outperformed the atlas-based MCA mask model internally (AUROC 0.88 vs. 0.82) and externally (AUROC 0.83 vs. 0.66).
• Incorporating CoW features further improved performance, reaching an AUROC of 0.87.

Conclusions:

• A fully automated and generalizable CTA radiomics pipeline for objective collateral scoring in acute LVO was developed.
• This approach enhances prediction accuracy and generalizability, supporting data-driven decision-making in stroke management.
• Automated scoring reduces inter-rater variability and improves workflow efficiency for individualized stroke treatment.