Researchers have developed a new method for extracting structured information from radiology reports. The approach, called FedPref, allows institutions with limited data to collaborate and improve the accuracy of extraction without sharing their actual reports or annotations. This is achieved through a federated learning process where local models are trained on shared model updates. Initial results show that FedPref outperforms traditional methods in terms of F1 scores, particularly for smaller hospitals. --- Why it matters: This matters to researchers and engineers working on AI applications in healthcare because it provides a more efficient way to extract structured information from radiology reports, which is essential for downstream search and analysis. The method can help institutions with limited data resources improve the accuracy of extraction without relying on large-scale centralized training datasets. Source: https://arxiv.org/abs/2608.16971