Google Research and Google DeepMind developed a Language Model (LLM) with conversational and collaborative capabilities to assist clinicians in diagnosing complex medical conditions. The LLM, fine-tuned on medical domain data, exceeded unassisted clinicians’ performance in providing accurate differential diagnoses (DDx). A study with 20 clinicians showed an accuracy of 59.1% with LLM assistance compared to 33.6% without. However, limitations were reported, such as the LLM’s inability to access tables and figures and difficulty in inputting case information. The LLM was found to be beneficial for simpler cases with specific keywords or signs. The researchers noted the potential for further real-world evaluation and the need for additional work to understand suitability for clinical settings.
Source link
Study: Google reveals LLM that helps accurately diagnosis complex cases
