<![CDATA[Background: Acute ischemic stroke due to large vessel occlusion (LVO) requires rapid identification. Reducing the time to diagnosis and treatment of stroke patients is an important goal to improve clinical outcomes. Non-contrast computed tomography (NCCT) is widely used in clinical practice for suspected stroke patients. Automated analysis using artificial intelligence in NCCT may be a solution to accelerate the early detection of LVO. Objective: To determine the accuracy of artificial intelligence in NCCT to predict LVO. Methods: A systematic literature search was conducted based on the PRISMA flow chart in four databases (PubMed, ProQuest, ScienceDirect, Cochrane Library) until June 2024. Data extraction was performed to evaluate the accuracy of predicting LVO. Quality assessment was performed using QUADAS-2. All data were analyzed using Review Manager 5.4 and MetaDTA 2.0. Results: Five studies involving 4.862 patients were enrolled. The quality of all the studies was high and had a low risk of bias. All studies used different software. Artificial intelligence in NCCT had fairly good accuracy with a sensitivity and specificity of 0.83 (95% CI; 0.78-0.87) and 0.73 (95% CI; 0.52-0.87). NCCT plus clinical status (NIHSS, stroke onset) in two studies slightly improved overall accuracy with a sensitivity and specificity of 0.85 (95% CI; 0.80-0.89) and 0.74 (95% CI; 0.54-0.88). Two studies reported that machine learning took less than two minutes. Conclusion: Artificial intelligence in NCCT was reasonably accurate and took a short time to predict LVO. There are still opportunities for machine learning to improve performance. Further research is still needed.]]>
