Statistical Shape Modeling and Prediction of Lumbar Spine Morphology in Patients With Adolescent Idiopathic Scoliosis.

Abstract

A lumbar spine statistical shape model (SSM) was developed to explain morphological differences in a population with adolescent idiopathic scoliosis (AIS). Computed tomography (CT) was used to collect data on the lumbar spine vertebrae and curvature of 49 subjects. The CT data were processed by segmentation, landmark identification, and template mesh mapping, and then SSMs of the individual vertebrae and entire lumbar spine were established using generalized Procrustes analysis and principal component analysis (PCA). Scaling was the most prevalent variation pattern. The weight coefficient was optimized using the Levenberg-Marquardt (LM) algorithm, and multiple regression analysis was used to establish a prediction model for age, sex, height, and body mass index (BMI). The effectiveness of the SSM and prediction model was quantified based on the root-mean-square error (RMSE). An automatic measurement method was developed to measure the anatomical parameters of the geometric model. The lumbar vertebrae size was significantly affected by height, sex, BMI, and age, with men having lower vertebral height than women. The trends in anatomical parameters were consistent with previous studies. The vertebral SSMs characterized the shape changes in the processes, while the lumbar spine SSM described alignment changes associated with translatory shifts, kyphosis, and scoliosis. Quantifying anatomical variation with SSMs can inform implant design and assist clinicians in diagnosing pathology and screening patients. Lumbar spine SSMs can also support biomechanical simulations of populations with AIS.