Identification of spinal deformity classification with total curvature analysis and artificial neural network

In this study, a multilayer feedforward, back-propagation Artificial Neural Network is implemented to identify the classification patterns of the scoliotic spinal deformity. At first step the simplified three-dimensional spine model is constructed from coronal and sagittal x-ray images. The features of the central axis curve of the spinal deformity patterns in 3D space are extracted by the Total Curvature analysis. The discrete form of the Total Curvature, including the curvature and the torsion of the central axis of the simplified 3D spine model is derived from the Difference Quotients. The values of Total Curvature of 17 vertebrae from first thoracic to the fifth lumbar spine formed a Euclidean space of 17 dimensions. Either the curvature or the torsion of the three-dimensional curve of the central axis of the spine model could provide the input of the artificial neural network. The King Classification model is tested on the neural network. Five sets of King spinal deformity patterns are randomly selected by the definition of King classification. The output layer of the artificial neural network has five neurons representing the five King classification types. The network validation was conducted by the Hold-Out method, one of Cross-Validation variant. The performance of the neural network is compared between two network topologies, one with one hidden layer and another with two hidden layers. The results are shown in a table with each of five datasets leave-out and all five datasets participating the training, with either one hidden layer or two hidden layer network.