Brief

Researchers have developed RadAgent, an AI system designed to improve the interpretation and reporting of chest CT scans. Unlike previous models that provide only final outputs, RadAgent generates reports through a stepwise, interpretable process, showing clinicians the intermediate decisions and tool interactions. This approach enhances clinical accuracy by 5.8 points in macro-F1 and 18.6% in micro-F1, significantly improves robustness under adversarial conditions, and introduces a new capability for faithfulness in reporting. AI

Researchers have introduced ASAP, a new pre-training framework designed to improve the learning of representations from medical volumetric scans like chest CTs. This framework incorporates anatomical knowledge and dynamically links textual findings from radiology reports to specific regions within the scans. ASAP has demonstrated state-of-the-art performance across a wide range of downstream tasks, particularly excelling in scenarios with limited supervision or distribution shifts. AI