Problems of assessing the clinical efficiency of artificial intelligence systemsin diagnosing ischemic stroke

Artificial intelligence (AI) is actively used in the diagnosis of ischemic stroke, allowing speeding up the decision-making process and improving the accuracy of diagnosis. Machine learning models are able to identify ischemic areas based on computed tomography and magnetic resonance imaging data, as well as indicate the volume of damage and calculate the ASPECTS score. Modern AI systems demonstrate high diagnostic accuracy, comparable to the accuracy of radiologists. According to clinical studies, these systems significantly reduce the time from the patient admission to the vascular center to treatment, but their impact on clinical outcomes remains unclear. The review discusses the problems of assessing the clinical effectiveness of AI in the diagnosis of ischemic stroke, including systematic bias in model training and choice of study design, as well as publication bias. To integrate AI into clinical practice, randomized controlled trials with clinically relevant endpoints, as well as standardization of data and methods for assessing effectiveness, are needed. Despite significant AI advances in diagnosing ischemic stroke, their effectiveness in real-world practice requires further study and validation.