Development of Spinal Deformity Detection System by Use of Moire Images

为了在早期检查脊柱畸形的存在，骨科医生传统上对儿童进行一种称为前屈测试的无痛检查。在前屈测试中，主要是医生检查一个肩膀是否比另一个肩膀低。但这种测试既不具有生殖性，也不客观。此外，在学校进行体检时，这种检查需要花费很多时间。为了克服这些困难，提出了一种云纹方法，该方法利用人体背部的云纹地形图像，通过视觉筛选以二维方式检查云纹图案的对称性/非对称性。本文提出了一种从云纹图像中自动检测脊柱畸形的新技术。在这项研究中，我们提出了从云纹地形图像中自动检测脊柱畸形的新技术。首先，将原始的云纹图像输入计算机，利用近似对称性分析在云纹图像上提取被试背部的中线；然后在云纹图像上从上到下自动选择感兴趣的区域。因此，不对称程度的数值表示在评估畸形时是有用的。计算云纹图像左、右侧区域相对于提取的中线的局部质心位移、密度特征的灰度值差和形状指标的肩角差。从左侧和右侧矩形区域提取统计特征向量用于训练神经网络和支持向量机。在实验中，取得了满意的分类效果。

In order to check the presence of spinal deformity in the early stage, orthopedists have traditionally performed on children a painless examination called a forward-bending test in school screening. In forward-bending test, mainly medical doctor checks to see if one shoulder is lower than the other. But this test is neither reproductive nor objective. Moreover, the inspection takes much time when applied to medical examination in schools. To overcome these difficulties, a moire method has been proposed which takes moire topographic images of human subject backs and checks symmetry/asymmetry of the moire patterns in a two-dimensional way on visual screening. In this paper, we propose a new technique for automatic detection of spinal deformity from moire topographic images.In this study, we propose new techniques for automatic detection of spinal deformity from moire topographic images. In the first stage, once the original moire image is fed into computer, the middle line of the subject's back is extracted on the moire image by employing the approximate symmetry analysis. Region of interest are then automatically selected on the moire image from its upper part to the lower part. Numerical representation of the degree of asymmetry is therefore useful in evaluating the deformity. Displacement of local centroids and difference of gray value from the density feature and angle of shoulders from the shape index are calculated between the left-hand side and the right-hand side regions of the moire images with respect to the extracted middle line. Extracted statistical feature vectors from the left-hand side and right-hand side rectangle areas apply to train the NN and SVMs. In the experimental, satisfactory classification results are achieved.