Work

Grpo-rad: Group Relative Policy Optimization for Radiology Report Summarization

Year: 2025

Type: conference-paper

Abstract: : We investigate Group Relative Policy Optimization (GRPO) for radiology report summarization using Qwen 3.0 models. GRPO enables optimization of composite reward functions combining syntactic and semantic measures, addressing limitations of traditional supervised fine-tuning. Our comprehensive evaluation on MIMIC-III demonstrates that GRPO consistently outperforms baseline and supervised fine-tuning approaches across multiple metrics including ROUGE-L and F1-RadGraph.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.17470608

Authors: Nassiri, Fargol, Rudzicz, Frank

Institutions: Dalhousie University, Vector Institute

Language: English

Topic: Topic Modeling

Subfield: Artificial Intelligence

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7102389599

FWCI:

Cites:

Cited by:

Related to: