The role of artificial intelligence in advancing scoliosis care: a rapid review of current evidence and future opportunities.

Abstract

<h4>Background</h4>Adolescent idiopathic scoliosis (AIS), is a complex three-dimensional deformity of the spine that affects a significant percentage of the adolescent population. Its progressive nature and evolution variability complicates therapeutic decisions, generating the need for more accurate tools for diagnosis, prediction of risk of progression and optimization of treatments. Artificial intelligence (AI) and machine learning (ML) emerge as tools with significant potential for comprehensive management of AIS. Despite the enthusiasm for these applications, there are important limitations that need to be addressed. The aim of this rapid review is to address a timely synthesis of available research and assess the quality of published reviews.<h4>Methods</h4>Systematic reviews and meta-analyses published by April 2025 in English on scoliosis and any intervention involving AI were included. Search was performed in Embase, Cochrane Review Database and Pubmed/medline using the terms MesH scoliosis idiopathic and intelligence artificial and systematic reviews and meta-analysis. Two independent reviewers screened titles and abstracts following the PRISMA RR guidelines and assessed full text articles with the AMSTAR 2 tool. Any disagreement was resolved by a third reviewer.<h4>Results</h4>Five systematic reviews met inclusion criteria. 55% of the included studies used AI algorithms with convolutional neural networks, artificial neural networks, decision trees, support vector machines, and hybrid models. The main applications in AIS were automatic Cobb angle measurement with high accuracy (< 3° in some models), curve type classification, prediction of curve progression and patient education and clinical decision support using language models.<h4>Conclusion</h4>AI offers promising solutions for AIS management, particularly in automated Cobb angle measurement and progression prediction. Combining deep learning models with clinical data may transform future practice, but external validation and clinical integration must be strengthened to enable effective implementation.