Large Language Model for Healthcare
Large language models (LLM) utilizing pre-trained generative transformer (GPT) architecture have achieved tremendous success in natural language processing (NLP). In healthcare applications, LLMs have also demonstrated great potential in processing and analyzing medical texts, paving the way for revolutionizing the healthcare industry, aiding physicians in their daily work, and improving the quality of patient care. At CAMCA, we have conducted a wide range of research on developing, improving, and harnessing LLM solutions for healthcare applications. Some of our recent healthcare LLM projects include: 1) RadiologyGPT, a fine-tuned LLAMA2 7B model using radiology reports from a public dataset; 2) ImpressionGPT, developing exemplar-based prompt engineering for summarizing radiology reports; 3) CohortGPT: establishing patient cohort via chain-of-thought technique on LLMs; 4) DeID-GPT: leveraging LLMs for removing HIPPIA identifiers from the medical text.
Related publications:
RadiologyGPT: https://arxiv.org/abs/2309.06419
ImpressionGPT: https://arxiv.org/abs/2304.08448
CohortGPT: https://arxiv.org/abs/2307.11346
DeID-GPT: https://arxiv.org/abs/2303.11032
https://link.springer.com/chapter/10.1007/978-3-031-43075-6_34
https://arxiv.org/abs/2307.13693
https://arxiv.org/abs/2306.10095
https://arxiv.org/abs/2306.08666
https://arxiv.org/abs/2305.03513
https://arxiv.org/abs/2304.11567
https://arxiv.org/abs/2304.09138
https://arxiv.org/abs/2304.01938
https://arxiv.org/abs/2302.10447
https://arxiv.org/abs/2302.13007
https://link.springer.com/chapter/10.1007/978-3-031-21014-3_28