Employing large language models safely and effectively as a practicing neurosurgeon

Abstract

BACKGROUND: Large Language Models (LLMs) have demonstrated significant capabilities to date in working with a neurosurgical knowledge-base and have the potential to enhance neurosurgical practice and education. However, their role in the clinical workspace is still being actively explored. As many neurosurgeons seek to incorporate this technology into their local practice environments, we explore pertinent questions about how to deploy these systems in a safe and efficacious manner. METHODS: The authors performed a literature search of LLM studies in neurosurgery in the PubMed database ("LLM" and "neurosurgery"). Papers were reviewed for LLM use cases, considerations taken for selection of specific LLMs, and challenges encountered, including processing of private health information. RESULTS: The authors provide a review of core principles underpinning model selection, including technical considerations such as model access, context windows, multimodality, retrieval-augmented generation, and benchmark performance, as well as relative advantages of current LLMs. Additionally, the authors discuss safety considerations and paths for institutional support in safe LLM inference on private health data. The resulting discussion forms a framework for key dimensions neurosurgeons employing LLMs should consider. CONCLUSIONS: LLMs present promising opportunities to advance neurosurgical practice, but their clinical adoption necessitates careful consideration of technical, ethical, and regulatory hurdles. By thoughtfully evaluating model selection, deployment approaches, and compliance requirements, neurosurgeons can leverage the benefits of LLMs while minimizing potential risks.

Ähnliche Arbeiten