A scoping review on the application of new technology in the screening and diagnosis of adolescent idiopathic scoliosis.

Abstract

Adolescent Idiopathic Scoliosis (AIS) affects 2-4% of adolescents globally. Current screening methods face challenges, including high false-positive rates, underscoring the need for safer, more accurate alternatives to improve early detection and reduce unnecessary imaging. This scoping review evaluates Artificial Intelligence (AI) applications in non-radiological AIS screening and diagnosis, aiming to (1) identify current AI models, (2) assess performance metrics, (3) examine validation methods, and (4) evaluate feasibility for large-scale implementation in current healthcare settings. Following PRISMA-ScR guidelines, 16 studies (2001-2025) were analyzed from PubMed, Scopus, and Web of Science. Inclusion criteria encompassed studies using AI with non-radiological inputs (RGB images, 3D scans, depth maps). Data on model types, performance metrics, validation, and clinical implications were synthesized. Convolutional Neural Networks (CNNs; 71.4%) predominated, followed by Support Vector Machines (21.4%) and Artificial Neural Networks (7.14%). AI models achieved a weighted mean accuracy of 91.1% (range: 72.4-97.5%) for curve classification and Cobb angle prediction errors of 5.5° ± 3.5°. For diagnosing scoliosis (curves > 10°), models achieved ≥ 87.5% sensitivity (up to 97.4% with advanced models) and specificity ≥ 83.5% (up to 99% with high-precision algorithms). AI-based non-radiological AIS screening demonstrates high diagnostic accuracy and offers a promising alternative to traditional methods by providing accessible, multi-output assessments using smartphone cameras or 3D scanners. Key limitations include small sample sizes, inconsistent validation, and ethical concerns (e.g., data privacy and automation bias). Future research should prioritize randomized controlled trials, standardized protocols, and multi-level stakeholder engagement to facilitate clinical integration.