Abstract 045: Emerging Trends and Future Directions in AI‐Driven Stroke Imaging: Bridging the Gap Between Technology and Clinical Practice

Introduction/Purpose Acute ischemic stroke (AIS) caused by large vessel occlusion (LVO) is a time‐critical condition requiring rapid diagnosis and intervention to optimize outcomes. Artificial intelligence (AI) has emerged as a transformative tool in stroke imaging, enhancing detection, triage, and outcome prediction. This abstract synthesizes recent advancements in AI‐driven stroke imaging, focusing on automated LVO detection, infarct volume prediction, and clinical workflow integration, to highlight emerging trends and future directions for bridging the gap between technological innovation and clinical practice. Materials and Methods A comprehensive review of 49 peer‐reviewed studies from 2019 to 2025 was conducted, focusing on AI applications in AIS imaging, specifically for LVO detection and outcome prediction using computed tomography angiography (CTA), CT perfusion (CTP), and other modalities. Studies were sourced from a provided dataset, including retrospective and prospective analyses from multicenter registries. Key metrics included sensitivity, specificity, area under the receiver operating characteristic curve (AUROC), Dice similarity coefficient (DSC), and workflow time reductions. AI tools such as convolutional neural networks (CNNs), deep learning models (e.g., Viz LVO, CINA LVO), and novel platforms (e.g., Openwater optical blood flow monitor) were evaluated. Statistical comparisons, including correlation coefficients and receiver operating characteristic (ROC) analyses, were used to assess AI performance against clinical standards. Results AI‐driven LVO detection demonstrated high sensitivity (74.6‐100%) and specificity (71‐97%) for proximal occlusions (ICA, M1), though sensitivity decreased for distal M2 occlusions (0‐58%). Automated platforms like Viz LVO and CINA LVO reduced door‐to‐notification times by 22‐25 minutes and improved workflow efficiency in hub‐and‐spoke models. Deep learning models using CTA predicted final infarct volumes with correlations up to r=0.92 and DSC of 0.35‐0.85, rivaling CTP‐based methods. Outcome prediction models integrating CTA and clinical variables achieved AUROCs of 0.81‐0.86 for 90‐day modified Rankin Scale (mRS) scores. Challenges included lower sensitivity for smaller lesions, limited model interpretability, and variability due to disease prevalence (4.1% vs. 45‐62% in training datasets), contributing to the accuracy paradox. Emerging tools, such as portable optical monitors, outperformed traditional scales (AUROC 0.82 vs. 0.65‐0.70), suggesting potential for prehospital triage. Conclusion AI‐driven stroke imaging has revolutionized LVO detection, infarct prediction, and workflow optimization, with significant reductions in treatment delays and improved prognostic accuracy. However, challenges such as variable performance for distal occlusions, data limitations, and the accuracy paradox highlight the need for tailored algorithms reflecting local disease prevalence. Future directions include integrating large language models, enhancing model interpretability, and developing large public datasets to improve generalizability. These advancements will bridge the gap between AI technology and clinical practice, enabling precision medicine and better patient outcomes in AIS management.