ARTIFICIAL INTELLIGENCE IN SCOLIOSIS DIAGNOSIS: A COMPARATIVE STUDY BETWEEN CHATGPT AND SURGEONS

Abstract

ABSTRACT Objective: This study explores the accuracy of ChatGPT in classifying and suggesting approaches for adolescent idiopathic scoliosis, assessing the level of agreement between the artificial intelligence model’s responses and the evaluations of spine surgery specialists. It aims to help answer the following question: Is it possible to trust ChatGPT-4 (natural language artificial intelligence) to recommend approaches for typical everyday cases, aiding less experienced orthopedists or even general practitioners? The proposed analysis seeks to identify the potential and limitations of applying artificial intelligence to support diagnosis and clinical decision-making without prior training of the platform. Methods: This is a cross-sectional study involving five fictitious cases of idiopathic scoliosis presented to ChatGPT, which provided the Lenke classification and a suggested approach for each case. A panel of 37 surgeons evaluated the responses, determined the best approach, and scored ChatGPT’s recommendations on a Likert scale from 1 to 5, reflecting their level of agreement. Results: In simpler cases (Case 1), ChatGPT showed high agreement with the specialists, with 97.3% of the surgeons agreeing with the recommendation of “instrumentation surgery” (AC1=0.95). However, agreement was significantly lower in more complex cases (Cases 3 and 5), with only 11.1% and 18.8% of the specialists accepting the Al’s recommendations, respectively. The model’s accuracy in the Lenke classification was consistent across all cases, demonstrating its ability to apply standardized criteria. There was no significant correlation between the surgeons’ experience and their level of agreement with the software. Conclusion: ChatGPT showed potential as an auxiliary tool in the diagnosis and therapeutic planning of scoliosis, particularly in classification, but it is not yet ready to be used reliably and consistently, especially in more complex cases, particularly when considering clinical nuances and individual patient factors. Although promising, the adoption of this technology can complement clinical judgment but still requires supervision and does not replace the role of specialized medical evaluation in the current scenario. Level of Evidence IV; Descriptive Observational Studies.

Ähnliche Arbeiten