Researchers at NIH found that an AI model excelled in solving medical quiz questions based on clinical images but struggled with explanations, highlighting the need for further evaluation before integrating AI into clinical practice.
- An AI model called GPT-4V showed high accuracy in answering medical quiz questions but often made mistakes in describing images and explaining diagnoses, according to researchers at the NIH.
- Physicians scored the AI model highly for selecting the correct diagnosis, with the AI model performing better than physicians in closed-book settings and physicians using open-book tools outperforming the AI model, especially on difficult questions.
- The AI model was less effective in recognizing relationships between different visual cues in images, demonstrating the importance of evaluating multi-modal AI technology before clinical use.
- The study used a multi-modal AI model called GPT-4V that combines text and images to enhance medical decision-making, highlighting the potential benefits and limitations of AI integration into healthcare practices.
- The research suggests that AI can complement human decision-making in healthcare by providing data-driven insights, but more studies are needed to compare AI models like GPT-4V with physicians’ diagnostic abilities.
Source link
Artificial Intelligence, Medical Decision-making, Internal Medicine, Nursing, Public Health & Prevention