基于贝叶斯网络的轻度认知障碍诊断系统

Diagnostic System of Mild Cognitive Impairment Based on Bayesian Network

提出一种基于依赖性分析和打分函数进行贝叶斯网络结构学习的新方法,并把该方法应用于轻度认知障碍诊断系统中。该算法首先通过对结点变量间的依赖性分析测试和无向图的遍历,获得贝叶斯网络结构中所有结点的先验顺序,然后用启发式打分—搜索方法获得最优的贝叶斯网络结构。实验结果表明,该算法能够在不增加算法复杂度的情况下,完成无结点顺序要求的贝叶斯网络学习,并能应用于轻度认知障碍诊断系统中,实现较好的预测,进而辅助医生的诊断。

This paper presents a new method in structure learning of Bayesian network based on dependency analysis and scoring function. Through analyzing the dependent relationship between variables and accessing to undirected graph, the prior sequence of all of the nodes in Bayesian network structure is obtained. The optimal structure of the Bayesian network is then generated by heuristic-search method. The new algorithm has been applied to the diagnostic system of mild cognitive impairment. The experimental results show that the new algorithm can better predict the possibility of mild cognitive impairment under the similar complexity, and further assist the diagnosis of doctor.