Prediction of Scoliosis Progression in Time Series Using a Hybrid Learning Technique

Abstract

Scoliosis is a common and poorly understood spinal disorder that is clinically monitored with a series of full spinal X-rays. The purpose of this study was to predict scoliosis future progression at 6- and 12-month intervals with successive spinal indices and a hybrid learning technique (i.e., the combination of fuzzy c-means clustering and artificial neural network (ANN)). Ultimately this could decrease scoliotic patients' radiation exposure and the associated cancer risk in growing adolescents. Seventy-two data sets were derived from a database of 56 acquisitions from 11 subjects (29.8 +/- 9.6 degrees Cobb angle, 11.4 +/- 2.4 yr), each consisting of 4 sequential values of Cobb angle and lateral deviations at apices in 6- and 12-month intervals in the coronal plane. Progression patterns in Cobb angles (n = 10) and lateral deviations (n = 8) were successfully identified using a fuzzy c-means clustering algorithm. The accuracies of the trained ANN, having a structure of three input variables, four nonlinear hidden nodes, and one linear output variable, for training and test data sets were within 3.64 degrees (+/- 2.58 degrees) and 4.40 degrees (+/- 1.86 degrees) of Cobb angles, and within 3.59 (+/-3.96) mm and 3.98 (+/- 3.41) mm of lateral deviations, respectively. Those results were twice the accuracy of typical clinical measurement (~10 degrees) and in close agreement with those using cubic spline extrapolation and adaptive neuro-fuzzy inference system (ANFIS) techniques. The adapted technique for predicting the scoliosis deformity progression holds significant promise for clinical applications.