面向多分类自闭症辅助诊断的标记分布学习

Label Distribution Learning for Computer Aided Diagnosis of Multi-class ASD Classification

自闭症谱性障碍(ASD)是一系列复杂的神经发展障碍性疾病,其包括若干与发育障碍相关的疾病,但是现有的自闭症辅助诊断方法大多是二分类方法,无法满足现实的需要。此外,ASD数据包含的标记噪声,以及高维度、数据分布不平衡等特点给传统分类方法带来了巨大的挑战。为此,提出一种新型的ASD辅助诊断方法,该方法通过引入标记分布学习(LDL)来解决标记噪声问题,引入代价敏感机制来解决样本不平衡问题,并采用基于支持向量回归(SVR)的标记分布学习方法,通过将样本映射到特征空间,解决高维特征带来的分类困难,最终实现多分类ASD的辅助诊断。实验结果表明,与已有方法比较,所提方法克服了多数类和少数类对结果的影响的不平衡性,可以有效地解决ASD诊断中的不平衡数据问题,拥有更好且稳定的分类性能,可以辅助ASD的诊断。

Autism spectrum disorder(ASD)is a series of complex neurodevelopmental disorders,including several diseases related to developmental disorders,but most of the existing diagnosis methods for autism are binary classification methods which cannot meet the actual needs.In addition,the label noise contained in ASD data,as well as the characteristics of high dimensionality and data imbalance,has brought great challenges to traditional methods.To this end,a new computer aided diagnosis method of ASD is proposed.This method solves the label noise by introducing label distribution learning(LDL),introduces a cost-sensitive mechanism to solve the data imbalance,uses label distribution support vector regression(SVR)to solve the classification difficulties caused by highdimensional features by mapping samples to the feature space,and finally realizes the computer aided diagnosis of multi-class ASD.Experimental results show that compared with the existing methods,the proposed method overcomes the imbalance of the influence of the majority class and the minority class on the results,can effectively solve the class imbalance in ASD diagnosis,and has better and stable classification performance,which can assist in the diagnosis of ASD.