We systematically investigate lightweight strategies to adapt large language models (LLMs) for the task of radiology report summarization (RRS). Specifically, we focus on domain adaptation via pretraining (on natural language, biomedical text, or clinical text) and via discrete prompting or parameter-efficient fine-tuning. Our results consistently achieve best performance by maximally adapting to the task via pretraining on clinical text and fine-tuning on RRS examples. Importantly, this method fine-tunes a mere 0.32% of parameters throughout the model, in contrast to end-to-end fine-tuning (100% of parameters). Additionally, we study the effect of in-context examples and out-of-distribution (OOD) training before concluding with a radiologist reader study and qualitative analysis. Our findings highlight the importance of domain adaptation in RRS and provide valuable insights toward developing effective natural language processing solutions for clinical tasks.
@article{arxiv.2305.01146,
title = {RadAdapt: Radiology Report Summarization via Lightweight Domain Adaptation of Large Language Models},
author = {Dave Van Veen and Cara Van Uden and Maayane Attias and Anuj Pareek and Christian Bluethgen and Malgorzata Polacin and Wah Chiu and Jean-Benoit Delbrouck and Juan Manuel Zambrano Chaves and Curtis P. Langlotz and Akshay S. Chaudhari and John Pauly},
journal= {arXiv preprint arXiv:2305.01146},
year = {2023}
}
Comments
12 pages, 10 figures. Published in ACL BioNLP. Compared to v1, v2 includes minor edits and one additional figure in the appendix. Compared to v2, v3 includes a link to the project's GitHub repository