High-Sensitivity Radiation-Free Triage for Adolescent Idiopathic Scoliosis via 3D Point Cloud Geometry

Abstract Widespread screening for Adolescent Idiopathic Scoliosis (AIS) is critical for early intervention, yet it is currently bottlenecked by the inherent limitations of traditional methods. Radiographic diagnosis poses cumulative radiation risks, while manual physical examinations are highly subjective and time-consuming. Recent non-invasive 2D computer vision approaches suffer from an unavoidable “dimensionality gap,” failing to capture critical depth and rotational information, which frequently leads to diagnostic misjudgments. To address these challenges, we present PointScol, a high-sensitivity, radiation-free triage system leveraging direct geometric processing of 3D back surface point clouds. Our framework employs a sequential pipeline: first, an automated segmentation module rigorously standardizes the input geometry by isolating the dorsal region of interest; subsequently, a diagnostic classification module evaluates the spinal deformity. Validation on a multi-center dataset (n=128) demonstrated that for the primary screening task (10° Cobb angle threshold), PointScol achieved 100.00% sensitivity in the external cohort, acting as a reliable gatekeeper to safely rule out healthy individuals without missing any cases requiring referral. Building upon the robust accuracy established at this 10° baseline, an extended 5-class grading module provides further diagnostic value. Rather than functioning as a rigid predictive task, this multi-class stratification acts as an advanced clinical assistant, offering nuanced severity insights to guide referral urgency and optimize medical resource allocation for high-risk patients. Collectively, this sequential design establishes PointScol as a safe and highly efficient clinical filter: it reliably prevents unnecessary radiation exposure for healthy adolescents while ensuring prioritized interventions for those most in need.