311 CNS+ Chatbot and AtlasGPT: Evaluating Domain-Specific LLMs in Neurosurgery

View at publisher Summarise with VIRAT
Authors: Ali A. Mohamed, Gabriel Flores-Milan, Emma C. Sargent, Mitch M. Johansen, Gersham J. Rainone, Elliot Pressman, Kunal V. Vakharia
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

INTRODUCTION: Large language models (LLMs) are increasingly being explored for their utility in neurosurgery. CNS+ chatbot and AtlasGPT are two neurosurgery-specific LLMs designed to support clinical reasoning, education, and other neurosurgery related tasks. Given their growing integration into clinical workflows, it is essential to understand their comparative performance across key neurosurgical functions. METHODS: A total of 113 prompts adapted from several sources were administered to CNS+ chatbot and AtlasGPT across seven domains: patient education, triage, challenging case management, operative note dictation, operative maneuver description, anatomy-based questions, and literature review. AtlasGPT responses were generated using “Patient style” for patient education and “Surgeon style” for all other tasks. Accuracy was rated on a 5-point scale. Domain-specific comparisons were analyzed for statistical significance. RESULTS: CNS+ chatbot scored significantly higher than AtlasGPT in challenging case management (5.0 vs 4.2, p = 0.022). It also slightly outperformed AtlasGPT in operative note generation, operative technique description, and triage, though differences were not statistically significant (4.4 vs 4.0, p = 0.098; 4.0 vs 3.6, p = 0.266; and 4.82 vs 4.61, p = 0.372, respectively). AtlasGPT outperformed CNS+ chatbot in literature review (5.0 vs 4.4, p = 0.123) and anatomy description (4.4 vs 4.0, p = 0.233), though these differences were also not statistically significant. For patient education, accuracy scores were identical overall (4.78 vs 4.78, p = 1.0) and across subdomains: spine (4.75 vs 4.75), cranial (4.6 vs 4.6), and general neurosurgery (5.0 vs 5.0). CONCLUSIONS: CNS+ chatbot and AtlasGPT demonstrated complementary strengths across neurosurgical tasks. CNS+ chatbot may be better suited for complex clinical reasoning, while AtlasGPT's comparable performance in literature review, anatomy description, and patient education supports its use in educational and documentation contexts.

Fields of Study & Subjects

Artificial Intelligence in Healthcare and Education Machine Learning in Healthcare Artificial Intelligence Applications

Publication details

Journal / source
Neurosurgery
Volume
72
Issue
Supplement_1
Pages
80–80
Publication date
26 March 2026
Language
EN