New PU-DPO framework tackles omission noise in radiology report generation

Researchers have developed a new framework called PU-DPO to address omission noise in vision-language models (VLMs) used for generating radiology reports. Standard training methods often lead these models to under-report findings because clinical reports themselves sometimes omit subtle details. PU-DPO reformulates the objective using positive-unlabeled learning, treating absent mentions as unlabeled rather than negative, and constructs preference supervision through edited model responses. This approach has shown consistent improvements in detecting and recovering findings across various pathologies in experiments. AI

IMPACT This research could lead to more accurate and reliable AI-generated radiology reports, improving diagnostic capabilities.

RANK_REASON The cluster contains a research paper detailing a new method for improving AI model performance. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• chest radiograph
• DagsHub
• Hugging Face
• Positive-unlabeled learning for disease gene identification
• PU-DPO
• vision-language model

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →