摘要
针对当前阿尔茨海默症脑皮层厚度数据的特征选择算法分类精度问题,提出一种融合的特征选择算法。分析处理轻度认知障碍人群和正常老年人的脑皮层厚度的核磁共振图像数据,基于此数据融合最小冗余和最大相关方法与Relief方法,并使用粒子群优化算法求得最优权重;使用此权重融合两种方法对脑皮层厚度的脑区特征进行特征选择,选出使分类准确率较高的特征。实验使用留一验证对实验结果进行评估,选出的特征对轻度认知障碍人群与正常老年人的分类效果好于当前流行的特征选择方法。
This paper proposed a hybrid feature selection algorithm,in view of the current problems of the classification accuracy of feature selection algorithm for cortical thickness data of Blzheimer＇s disease. Firstly,this paper analyzed and processed magnetic resonance image cortical thickness＇ data of mild cognitive impairment and normal elderly population. Based on this data it fused minimum redundancy and maximum correlation method and Relief method,and used particle swarm optimization algorithm to obtain the optimal weights. Then,it did feature selection using this weight fusion of two kinds of methods of cortical thickness for features of brain regions,selected characteristics that made the higher classification accuracy. The experiment uses leave-one-out cross validation to evaluat the result. The classification results of people with mild cognitive impairment and normal elderly people based on the selected features is better than the current popular feature selection method.
出处
《计算机应用研究》
CSCD
北大核心
2017年第3期675-677,682,共4页
Application Research of Computers
基金
国家自然科学基金资助项目(61210010)
关键词
阿尔茨海默症
轻度认知障碍
大脑皮层厚度
特征选择
Alzheimer＇s disease
mild cognitive impairment
cerebral cortex thickness
feature selection
