RadSearch, a domain-specific semantic search model, has been developed to retrieve radiology reports with unparalleled precision and enhance large language model diagnostic performance. Trained using contrastive learning on structured radiology reports, RadSearch outperforms existing models in retrieving relevant findings and improving diagnostic accuracy. This tool offers functionalities for finding similar cases and providing differential diagnoses, which could significantly impact radiologists' ability to enhance clinical decision-making and streamline workflows.

Source: Radiology