Artificial Intelligence-Based Virtual Assistant for the Diagnostic Approach of Chronic Ataxias

Area of Science:

• Neurology
• Artificial Intelligence
• Medical Diagnostics

Background:

• Chronic ataxias encompass over 300 heterogeneous diseases, presenting significant diagnostic challenges.
• Clinical and genetic variability complicates the accurate identification of these conditions.
• Artificial intelligence (AI) offers a potential solution for improving the diagnosis of chronic ataxias.

Purpose of the Study:

• To develop and validate an AI-powered virtual assistant for diagnosing chronic ataxias.
• To assess the diagnostic accuracy and usability of the AI assistant.
• To compare the AI assistant's performance against human specialists and existing AI models.

Main Methods:

• A virtual assistant was created using advanced algorithms, decision trees, and large language models.
• Validation involved analyzing 453 literature cases from 151 distinct causes of chronic ataxia.
• Diagnostic accuracy, time efficiency, and question count were compared with 21 neurologists and GPT-4.

Main Results:

• The AI virtual assistant achieved a 90.9% diagnostic accuracy, significantly surpassing neurologists (18.3%) and GPT-4 (19.4%).
• The AI tool demonstrated superior performance across various ataxia classifications, including age distribution, inheritance patterns, and clinical manifestations.
• The AI assistant required fewer questions (14) and less time (1.5 minutes) compared to neurologists (19.4 minutes).

Conclusions:

• The AI virtual assistant is an accurate, fast, and user-friendly tool for diagnosing chronic ataxias.
• This AI approach can serve as a valuable support tool in neurological consultations.
• The diagnostic framework has the potential for broader application to other neurological and non-neurological conditions.