AI Improves Non-Contrast CT Interpretation for Time-Sensitive Stroke Assessment

Emergency departments often use computed tomography angiography to identify these blockages, but contrast requirements and workflow constraints can delay imaging, potentially leaving large vessel occlusions undetected within critical treatment windows. To support earlier detection, researchers have validated an artificial intelligence tool that identifies large vessel occlusion on standard non-contrast CT.

The non-contrast CT-based large vessel occlusion detection algorithm from JLK was validated by investigators at Korea University Guro Hospital, Korea University College of Medicine, with collaborators at Seoul National University Bundang Hospital. The system analyzes routine head scans to flag suspected large vessel occlusion without the need for contrast imaging. Findings were published in the Journal of NeuroInterventional Surgery.

The team evaluated diagnostic performance using a multinational cohort of 963 patients, including 723 from Korea and 240 from the United States. Standalone area under the receiver operating characteristic curve (AUC) reached 0.963 in the Korean cohort and 0.899 in the U.S. cohort. High diagnostic performance persisted across different countries and various computed tomography scanners.

In a separate reader study involving eight clinicians, including specialists and residents, AI support improved interpretation of non-contrast CT. Reader accuracy increased from an AUC of 0.718 without support to 0.852 with support. Sensitivity rose from 46.6% to 63.7%, and specificity increased from 91.9% to 94.9%, equating to one additional large vessel occlusion identified for every 18 scans reviewed with assistance. No clear evidence of automation bias was observed.

Because endovascular thrombectomy is crucial to reopen blocked vessels, early recognition of large vessel occlusion directly influences prognosis. The results indicate that artificial intelligence can serve as a safety net to complement clinician judgment during time-sensitive stroke care. The approach addresses delays or barriers to computed tomography angiography by leveraging the most accessible imaging available in the emergency setting.

"This is clinically very significant: We can now rapidly and accurately identify severe stroke patients in the time-sensitive emergency room using only the most accessible equipment: non-contrast CT," said Chi Kyung Kim, Department of Neurology, Korea University Guro Hospital.

"With the help of AI, inexperienced medical staff or resident doctors can achieve diagnostic accuracy close to that of skilled specialists. This will significantly raise the standard of care at local hospitals and during nighttime emergencies," said Jun Sun-woo, Department of Radiology, Seoul National University Bundang Hospital.

Related Links
Korea University College of Medicine
Korea University Guro Hospital