FedPref: Federated Preference Learning for Structured Radiology Report Extraction

2026-08-17

AI Summary

"This study addresses the challenge of utilizing free text in radiology reports and the issue of non-uniform data distribution across different healthcare institutions, where direct data sharing is not feasible. It proposes a federated preference learning approach that leverages a frozen public language model for extracting information in JSON format. By collaboratively training compact Qwen3-8B adapters and sharing only model updates, the method mitigates the risk of collapse due to sample repetition from a single model, using a heterogeneous teacher pool. Experiments on six simulated hospital datasets show an average improvement of 2.49 points in F1 score, with the worst-performing site gaining 9.10 points. The approach achieved an F1 score of 68.68 on a manually validated test set of 400 reports."

Abstract

Radiology reports describe findings and locations in free text, but downstream search and analysis require these relations in a fixed schema. Learning this extraction requires labels that are unevenly distributed across institutions: smaller hospitals have less local evidence, and pooling data may be infeasible. We introduce FedPref: frozen public language models propose alternative JSON extractions, local annotations rank them, and sites collaboratively train compact Qwen3-8B adapters while sharing only model updates. A heterogeneous teacher pool provides cross-model contrast when repeated single-model samples collapse. On development data from six simulated hospitals with unequal data volume and disease prevalence, FedPref improves client-mean F1 by 2.49 points and worst-site F1 by 9.10 points compared with training each site in isolation, with the largest gains at the sites holding the least data. Central training on the pooled preference-pair union is 2.66 points higher on client-mean F1. On a locked, 400-report manually validated gold test set, FedPref reaches 68.68 F1 and pooled training 71.67, preserving that same ordering. FedPref thus lets institutions with unequal, unpooled data benefit from collaboration without ever sharing reports or annotations.

Problem

• Radiology Reports
• Structured Extraction
• Federated Learning
• Data Imbalance
• Collaborative Training

Innovation

• Federated Preference Learning
• Frozen Public Language Models
• Compact Qwen3-8B Adapters
• Cross-Model Contrast
• Data Privacy