Machine Learning Prediction of Disease Progression in Adolescent Idiopathic Scoliosis

Overview

Background and Problem Overview Adolescent Idiopathic Scoliosis (AIS) is a progressive musculoskeletal disorder characterized by a three-dimensional deformation of the spine occurring during adolescence. Diagnosis is typically established with a Cobb angle exceeding 10° and the presence of axial rotation. While the exact etiology remains unknown, leading theories include tissue abnormalities (muscle fibers, bone volume), impaired spinal biomechanics (asymmetric bone growth), and neurological factors (asymmetric cortical thickness, cerebral lateralization, and body schema distortions). The progressive nature of AIS, particularly the high risk of advancement at the onset of puberty, complicates clinical decision-making. Treatment is traditionally divided into three stages: Observation and Exercise: For Cobb angles between 10°-25°. Exercise and Bracing: For Cobb angles between 25°-45°. Surgery: For Cobb angles exceeding 45°. Despite these guidelines, the unpredictable progression of the disease and difficulties in treatment adherence create significant dilemmas. Specifically, for cases on the borderline of surgical indication, clinicians face the challenge of choosing between immediate surgery or conservative monitoring. Currently, there is no definitive method to predict progression, and patients are typically monitored in 6-month intervals. During these intervals, a patient's condition may remain stable or deteriorate significantly. Furthermore, guidelines recommend wearing a brace for an average of 18 hours per day, often for several years. This requirement is physically and psychologically demanding for adolescents, leading to poor compliance due to aesthetic concerns, functional limitations, and skin irritation. The inability to predict progression often leads to overtreatment (unnecessary bracing) or undertreatment (delayed intervention), both of which pose risks to the patient's long-term health. Radiological Concerns Disease progression is monitored via direct radiography (X-rays). However, frequent imaging increases the lifetime risk of cancer due to cumulative ionizing radiation. Notably, the risk of breast cancer in girls with AIS is reported to be approximately seven times higher than in the healthy population. Conversely, extending follow-up intervals risks missing windows for early intervention. An artificial intelligence (AI) model capable of predicting curve progression could optimize imaging frequency, ensuring safety while maintaining clinical efficacy. Objective and Methodology of the Study The primary aim of this research is to develop a machine learning-based model to predict the Cobb angle following a 12-week exercise intervention. The model will utilize comprehensive baseline and post-treatment data, including: Demographic and Anthropometric Data (Age, height, weight, gender). Clinical Assessments (Cobb angle, Risser score, angle of trunk rotation). Functional and Physical Metrics (Trunk muscle strength, Maximal Inspiratory and Expiratory Pressure \[MIP/MEP\], Biodex balance measurements). Visual Assessments (Walter Reed Visual Deformity Scale \[WRVAS\]). Research Hypotheses Primary Hypothesis: A machine learning model trained on pre- and post-exercise assessment data can significantly predict the Cobb angle at the end of a 12-week period with both statistical and clinical accuracy. Secondary Hypothesis: By predicting the risk of progression (the probability of an increase in Cobb angle), this model will contribute to reducing unnecessary surgical interventions, overtreatment (bracing/surgery), and cumulative X-ray exposure.