Zastosowanie sztucznej inteligencji do polepszenia diagnostyki epilepsji

OBJECTIVE: To assess the efficacy of an AI-driven History of Present Illness (HPI) tool in improving the diagnostic accuracy of epilepsy. BACKGROUND: Diagnosing epilepsy remains a significant challenge, particularly for non-neurologists and primary care providers. The median diagnostic delay for patients with new-onset focal epilepsy is estimated at 12 months. This study investigates whether an AI agent can improve diagnosis by standardizing the collection of clinical history. METHODS: We conducted a retrospective analysis of de-identified histories from 40 consecutive patients admitted to an inpatient Epilepsy Monitoring Unit. The quality of histories recorded by neurology residents was evaluated by a blinded, independent, and experienced epileptologist. We developed a web-based AI agent incorporating critical diagnostic questions. The AI Epilepsy HPI tool was used to process the clinical data, and its results were compared against video EEG Monitoring, the "gold standard" of epilepsy diagnosis. RESULTS: The independent epileptologist determined that the HPIs obtained by residents were adequate for diagnosis in only 58% of cases. Consequently, the expert's diagnostic accuracy based on those HPIs alone was also 58%. In contrast, the AI Epilepsy HPI tool achieved a diagnostic accuracy of 90% when compared to the video-EEG results. Also, the expert diagnosis was improved by the AI tool to 80% accuracy. SIGNIFICANCE: This study demonstrates that an AI agent can effectively assist providers in the systematic collection of clinical symptomatology for epilepsy. By standardizing the history-taking process, this tool has the potential to improve the education of junior neurology residents.