On Automatic Scoliosis Detection using a Neural Network

Abstract

In this paper, we propose a technique for automatic scoliosis detection method from moiré topographic images. Normally the moiré stripes show a symmetric pattern as a human body is almost symmetric. According to the progress of the deformity of a spine, asymmetry becomes larger. Numerical representation of the degree of asymmetry is therefore useful in evaluating the deformity. First, displacement of local centroids is evaluated statistically between the left-hand side and the right-hand side regions of the moiré images with respect to the extracted middle line. The degree of the displacement with respect to normal and abnormal cases is learned by a neural network employing the BP algorithm. An experiment was performed employing 1,200 moiré images (600 normal, 600 abnormal) and 89% of the images were classified correctly by the NN.