Related Experiment Video
Updated: Jan 9, 2026

The Adventures of Fundi Intervention Based on the Cognitive and Emotional Processing in Attention Deficit Hyperactive Disorder Patients
Published on: June 12, 2020
A multi-agent approach to neurological clinical reasoning
Moran Sorka1,2, Alon Gorenshtein1,3, Dvir Aran2,4
1AI Neurology Laboratory, Ruth and Bruce Rapaport Faculty of Medicine, Technion-Institute of Technology, Haifa, Israel.
Large language models show promise in neurology, but a novel multi-agent system significantly improved performance on complex clinical reasoning tasks, outperforming retrieval-augmented generation.
Area of Science:
- Artificial Intelligence in Medicine
- Clinical Neurology Decision Support
- Large Language Model Evaluation
Background:
- Large language models (LLMs) demonstrate potential in medicine, but their specialized reasoning for clinical neurology requires systematic assessment.
- Neurological assessment involves complex cognitive processes like anatomical localization and temporal pattern recognition, crucial for board certification.
Purpose of the Study:
- To systematically evaluate the performance of various large language models (LLMs) on specialized neurological reasoning tasks.
- To compare the effectiveness of base LLMs, retrieval-augmented generation (RAG), and a novel multi-agent system for neurological assessment.
Main Methods:
- Developed a benchmark of 305 questions from Israeli Neurology Board Certification Exams, categorized by complexity.
- Evaluated ten LLMs, including base models, RAG-enhanced versions, and a multi-agent system designed to mimic specialized cognitive functions.
- Conducted external validation using 155 neurological cases from the MedQA dataset.
Main Results:
- OpenAI-o1 achieved the highest base performance (90.9%); specialized medical LLMs performed poorly (Meditron-70B at 52.9%).
- Retrieval-augmented generation (RAG) offered variable improvements, with significant gains for mid-tier models but limited impact on high-complexity questions.
- The multi-agent framework dramatically improved performance, especially for mid-range models (LLaMA 3.3-70B reached 89.2% accuracy), particularly on complex questions.
Conclusions:
- A structured multi-agent approach significantly enhances complex medical reasoning in neurology, outperforming RAG and base LLMs.
- The multi-agent system effectively addresses persistent neurological reasoning challenges, transforming inconsistent performance into uniform excellence.
- This approach offers promising directions for AI assistance in challenging clinical neurological contexts.
Related Concept Videos
Critical Thinking II
Reason and Intuition
Patient-centered Care
Critical Thinking I
Introduction to Cognitive Psychology
This field emerged in the mid-20th century, following a period dominated by behaviorism, which...
Cognitivism
Previously dominated by behaviorism, which prioritized observable behaviors and largely ignored mental processes, psychology transformed in the 1950s. Cognitive psychologists argue that understanding how we think and process...

