Data-driven Classification of the 3D Spinal Curve in Adolescent Idiopathic Scoliosis with an Applications in Surgical Outcome Prediction

Adolescent idiopathic scoliosis (AIS) is a three-dimensional (3D) deformity of the spinal column. For progressive deformities in AIS, the spinal fusion surgery aims to correct and stabilize the deformity; however, common surgical planning approaches based on the 2D X-rays and subjective surgical decision-making have been challenged by poor clinical outcomes. As the suboptimal surgical outcomes can significantly impact the cost, risk of revision surgery, and long-term rehabilitation of adolescent patients, objective patient-specific models that predict the outcome of different treatment scenarios are in high demand. 3D classification of the spinal curvature and identifying the key surgical parameters influencing the outcomes are required for such models. Here, we show that K-means clustering of the isotropically scaled 3D spinal curves provides an effective, data-driven method for classification of patients. We further propose, and evaluate in 67 right thoracic AIS patients, that by knowing the patients’ pre-operative and early post-operation clusters and the vertebral levels which were instrumented during the surgery, the two-year outcome cluster can be determined. This framework, once applied to a larger heterogeneous patient dataset, can further isolate the key surgeon-modifiable parameters and eventually lead to a patient-specific predictive model based on a limited number of factors determinable prior to surgery.