Introduction and validation of OSCAR-optimal stent choice algorithm

Standardization and international guidelines for stent size selection are lacking. In this study, we introduce and validate an artificial intelligence (AI)- and image processing-supported, modular software algorithm trained on a multicentric vascular segmentation dataset that identifies stenoses, performs segmentations of stenotic vessel segments and suggests the optimal stent size for implantation.

This retrospective multicenter study included 149 patients who underwent stent implantation for symptomatic stenoses of the common and external iliac arteries between August 2017 and July 2024. Peri-interventional angiography datasets were evaluated by four board-certified interventional radiologists. For AI-training, all relevant stenoses were annotated and segmented to reflect intended stent sizing. The segmentation criteria were consensus-defined, and all readers completed a prior training session to ensure consistency. The modular algorithm comprises components for stenosis detection, segmentation and stent parameter prediction. Following pre-training on a publicly available coronary artery dataset, the model was fine-tuned on the study-specific iliac artery dataset using leave-one-out cross-validation.

OSCAR detected stenoses in 84.6% of cases. The model achieved a high recall (0.89 ± 0.21), meaning that most expert-annotated stenoses were correctly identified, while a moderate precision (0.65 ± 0.28) indicated some false-positive detections. Segmentation accuracy was good (DSC 0.77 ± 0.11). Stent diameter and length predictions demonstrated mean absolute percentage errors of 0.13 ± 0.18 and 0.33 ± 0.31, respectively, comparable to expert variability.

This proof-of-concept study demonstrates the potential of AI-assisted stent selection in vascular interventions. Furthermore, the option of a closed-loop framework promotes sustainability, reproducibility and cost-effectiveness in stent implantation procedures.