<< Back to the abstract archive
A Novel Plastic Surgery Literature Specific Artificial Intelligence Large Language Model: A Proof-of Concept Study With GPT-2
Berk B. Ozmen, Graham Schwarz
Cleveland Clinic
2024-02-01
Presenter: Berk B. Ozmen
Affidavit:
Yes. All original work of the authors.
Director Name: Graham Schwarz
Author Category: Fellow Plastic Surgery
Presentation Category: Clinical
Abstract Category: General Reconstruction
BACKGROUND: Plastic surgery literature demands innovative strategies to synthesize relevant literature for training and clinical use. Recently popular ChatGPT and artificial intelligence large language models offer an innovative solution. In this proof-of-concept study, we aimed to develop a large language model specific to plastic surgery, acting like a ChatGPT based on the current plastic surgery literature.
METHODS: We developed a Python code to automate the collection of 100 plastic surgery abstracts from PubMed. The abstracts were then preprocessed, tokenized, and formatted in Python and TensorFlow ecosystem into a structure suitable for model training. We employed the pre-trained GPT-2 model, which was originally developed and open-sourced by OpenAI. The model was fine-tuned on our dataset in a GPU- enabled environment, adjusting its parameters to better align with our plastic surgery specific data. We evaluated the performance of the resulting model based on its ability to generate coherent and contextually accurate text related to plastic surgery when prompted.
RESULTS: The resulting model demonstrated promising proficiency in generating plastic surgery related text, suggesting the successful integration of the plastic surgery domain specific knowledge from the fine-tuning process.
CONCLUSION: Our proof-of-concept study highlights the potential of large language models, like GPT-2, in plastic surgery when fine-tuned with domain-specific literature. Future works can focus on the model's potential in generating coherent hypotheses, summarizing literature, and explaining complex plastic surgery concepts. The Python code we developed for this task promises innovative implementations of large language models and artificial intelligence in plastic surgery.
