Artificial Intelligence-Driven Detection of Large Vessel Occlusions on NCCT: A Multi-Institutional Study

Area of Science:

• Neurology
• Radiology
• Artificial Intelligence in Medicine

Background:

• Perfusion imaging is standard for stroke patient triage.
• Non-contrast CT (NCCT) based triage for stroke is limited.
• Identifying anterior circulation large vessel occlusions (LVO) is critical for acute ischemic stroke (AIS) management.

Purpose of the Study:

• To evaluate the predictive capability of the "Triage Stroke" deep learning algorithm for identifying LVO on NCCT.
• To assess the algorithm's performance in patients with suspected AIS.

Main Methods:

• A multi-institutional study involving 612 patients with suspected AIS.
• Analysis of a balanced cohort with and without anterior circulation LVO.
• Ground truth established by concurrent CT angiography (CTA) evaluated by neuroradiologists.

Main Results:

• Triage Stroke achieved 67% sensitivity and 93% specificity for LVO detection on NCCT (AUC 0.8).
• The algorithm significantly outperformed general and subspecialty radiologists in LVO detection (P < .001).
• Incorporating NIHSS improved specificity to 99% and positive predictive value to 91%.

Conclusions:

• Triage Stroke demonstrates strong predictive accuracy for NCCT-based LVO detection, exceeding radiologist performance.
• The algorithm, especially when combined with NIHSS, can simplify the identification of endovascular treatment candidates.
• This technology offers potential benefits for stroke triage in resource-limited settings globally.