基于SVM的骨质疏松识别算法

Classification of osteoporosis by SVM based algorithm

文中讨论了基于模式分类的算法,通过常规的体检参数对骨质疏松情况进行预测和识别。由于常规体检参数和骨质疏松诊断结果之间的线性相关度小、参数方差大等问题,基于线性分类边界模型得到的分类器误差大,文中利用数据和骨质疏松之间的非线性关联特性,使用高斯核函数将原始训练数据映射到核空间进行分类,较好地实现了用体检参数预测骨质疏松。此外文中给出了利用多个分类器的分类结果进行组合方法,使得不同分类器分类结果相互矛盾时能够输出唯一的诊断结论。

This paper discusses on the classification algorithm of osteoporosis from data collected by normal physical examination. Due to the low correlation coefficients between osteoporosis and data from normal physical examination and also the large covariance of physical examination data, traditional linear methods such as the linear regression cannot be used. It makes use of the nonlinear proper of data, and builds the classification algorithm on kernel Hilbert space. With the Gaussian Radial basis function, the osteoporosis can be classified with low error. Besides the design of classification algorithm, it proposes information combination method based on the result of multiple classification algorithms so that a unique conclusion can be drawn when outputs from different classificatory contradict to each other.