PromptRad: AI Method for Low-Resource Radiology Report Labeling
A team of researchers has introduced PromptRad, a multi-label prompt-tuning technique that enhances knowledge for labeling radiology reports in low-resource environments. This innovative method treats multi-label classification as masked language modeling and utilizes synonyms from the UMLS Metathesaurus within a multi-word verbalizer to improve category representation. By fine-tuning a pre-existing language model without the need for extra classification layers, PromptRad significantly reduces the amount of labeled data required compared to traditional fine-tuning methods. Tests conducted on liver CT reports highlight its efficacy, tackling the issues posed by varied clinical descriptions and scarce labeled data in medical contexts.
Key facts
- PromptRad is a knowledge-enhanced multi-label prompt-tuning approach for radiology report labeling.
- It reformulates multi-label classification as masked language modeling.
- It incorporates synonyms from the UMLS Metathesaurus into a multi-word verbalizer.
- It fine-tunes PLMs without additional classification layers.
- It requires less labeled data than conventional fine-tuning.
- Experiments were conducted on liver CT reports.
- The method targets low-resource clinical settings.
- It addresses diverse descriptions in clinical reports.
Entities
Institutions
- UMLS Metathesaurus