Use of Artificial Intelligence in Screening for Adolescent Idiopathic Scoliosis: A Scoping Review

Adolescent idiopathic scoliosis (AIS) screening remains controversial, particularly regarding its effectiveness and cost-effectiveness in population settings. The Scoliosis Research Society - AIS screening (SRS-AIS) is simple, reliable, and widely accepted in clinical practice. With the emerging of artificial intelligence (AI) in medical screening, AI-based approaches have been proposed as potential alternatives. However, whether AI-AIS screening offers meaningful advantages over established SRS-AIS screening in real-world practice remains unclear. This study aimed to evaluate the efficacy, methodological quality, and practical relevance of AI-based AIS screening in comparison with conventional SRS-AIS screening approache.

A scoping review was conducted following PRISMA-ScR guidelines. PubMed, Scopus, and Web of Science were searched from inception to October 2025 for studies applying AI-based AIS screening. Extracted data included study design, participant characteristics, screening modality, model type, reference standard, and diagnostic performance metrics. Studies were classified as true screening (population-based, community settings) or quasi-screening (clinical datasets without population-level implementation). We evaluated whether studies reported direct comparisons with SRS-AIS screening, validation against standardized SRS-AIS performance, workflow feasibility, or cost-effectiveness analyses. Methodological quality was assessed using PROBAST-AI and QUADAS-AI.

Twenty-six studies published between 2018 and 2025, involving more than 330,000 participants across 12 countries, met inclusion criteria. Fourteen studies (54%) were true screening investigations and twelve (46%) were quasi-screening validations. Screening modalities included back-image or depth-camera systems (n=12), radiographic evaluations (n=5), ultrasound (n=3), census models (n=3), surface topography (n=2), and gait analysis (n=1). Median area under the curve was 0.94 (IQR 0.90-0.98), mean accuracy was 0.91, and mean Cobb angle error was 3.1°. Only eight studies were multicenter and seven were prospective. No study directly compared AI-AIS screening with SRS-AIS screening, and no cost-effectiveness analyses were identified. PROBAST-AI rated 62% of studies as low risk of bias, 35% as moderate, and 4% as high risk; QUADAS-AI classified 73% as low risk.

AI-AIS screening demonstrates technical feasibility and promising diagnostic performance; however, current evidence does not support replacing established SRS-AIS screening in routine practice. Major evidence gaps include the absence of direct comparative studies, economic evaluations, and prospective multicenter validation. These limitations must be addressed before AI-AIS screening can be recommended for widespread clinical or population-level implementation.