Case scenario generators for trauma surgery simulation utilizing autoregressive language models.

Area of Science:

• Medical Education
• Artificial Intelligence in Healthcare
• Trauma Surgery

Background:

• Trauma is a leading cause of death, necessitating advanced clinical expertise.
• The Advanced Trauma Life Support (ATLS) protocol requires extensive clinical experience, often gained through exposure to varied patient cases.
• Current methods for creating training case scenarios are time-consuming and require significant domain expertise.

Purpose of the Study:

• To develop an AI-powered tool for generating realistic trauma case scenarios.
• To provide accessible clinical experience to trainees without patient harm.
• To overcome the limitations of traditional case scenario authorship.

Main Methods:

• Developed Trauma AI, utilizing an autoregressive generative model (GPT2) based on the transformer architecture.
• Trained the model on 1.1 million case scenarios from the National Trauma Data Bank (NTDB).
• Integrated an out-of-domain detection mechanism to filter unrealistic scenarios, enhancing realism.

Main Results:

• Trauma AI successfully generated realistic case scenarios encoding the ATLS protocol.
• The model created novel scenarios not present in the original dataset.
• Unsupervised filtering of out-of-domain sequences significantly improved the realism of generated scenarios.

Conclusions:

• AI, specifically autoregressive models trained on large datasets, offers a viable solution for generating diverse and realistic trauma case scenarios.
• Trauma AI can enhance clinical training by providing broad exposure to injury permutations and rare scenarios.
• The developed filtering method effectively improves the quality and plausibility of AI-generated medical training materials.