Automatic Summarization of Doctor-Patient Encounter Dialogues Using Large Language Model through Prompt Tuning

The study introduces a method for automatic summarization of doctor-patient dialogues using the GatorTronGPT model, which is a generative clinical LLM with up to 20 billion parameters trained on 277 billion words. It employs prompt-tuning strategies that do not require updating model parameters, resulting in low computational costs. Experimental results indicate that GatorTronGPT-20B outperforms a fine-tuned T5 model on the MTS-DIALOG clinical benchmark, highlighting its effectiveness for practitioners seeking efficient solutions in clinical text summarization.