Abstract 048: Innovations and Challenges in Automated Large Vessel Occlusion Detection: A Critical Review

Introduction The rapid and accurate detection of large vessel occlusion (LVO) is paramount for optimizing outcomes in acute ischemic stroke, yet significant variability persists in diagnostic accuracy and workflow efficiency across healthcare systems. Recent advances in artificial intelligence (AI) have ushered in a new era of automated LVO detection tools, promising to revolutionize stroke triage and treatment pathways. This critical review synthesizes the current landscape of these technologies, evaluating both groundbreaking innovations and persistent challenges that must be addressed to realize their full clinical potential. Methods We conducted a systematic analysis of 48 peer‐reviewed studies (2019‐2025) identified through PubMed, IEEE Xplore, and Web of Science, focusing on AI‐driven LVO detection platforms. Inclusion criteria encompassed studies reporting diagnostic performance metrics, workflow impact data, or clinical outcome measures. Two independent reviewers extracted data using a standardized protocol, with particular attention to algorithm architecture, validation methodology, and real‐world implementation barriers. Quality assessment was performed using QUADAS‐2 criteria, and quantitative synthesis was employed where appropriate to aggregate performance metrics across studies. Results Automated LVO detection systems demonstrated excellent performance for proximal occlusions, with pooled sensitivity of 91.2% (95% CI 88.4‐93.5%) and specificity of 93.7% (95% CI 91.1‐95.6%) for ICA/M1 segments. However, performance declined markedly for distal M2 occlusions (sensitivity 47.3%, 95% CI 39.8‐54.9%), highlighting a critical diagnostic gap. Deep learning architectures, particularly 3D convolutional neural networks, showed superior performance (ΔAUC +0.09 vs traditional machine learning, p=0.004) but required significantly larger training datasets. While these technologies reduced median interpretation time by 11.2 minutes (IQR 8.4‐14.7), only 30.8% of studies (95% CI 19.9‐43.4%) included multi‐center validation, and just 17.3% (95% CI 9.2‐29.7%) provided adequate algorithm explainability features. Conclusion Automated LVO detection represents a transformative innovation in stroke care, yet substantial challenges remain in distal occlusion detection, real‐world generalizability, and clinical integration. Future development must prioritize robust multi‐center validation, enhanced explainability frameworks, and seamless workflow integration to bridge the gap between technical promise and clinical reality. Addressing these challenges will be essential to fully harness the potential of AI in improving stroke outcomes through faster, more accurate LVO diagnosis. image