脑皮层形态学多特征融合的轻度认知障碍的分类研究

Classification Study of Mild Cognitive Impairment Based on Multi-feature Fusion for Cortical Morphology

轻度认知障碍是介于老年痴呆与正常衰老之间的中间状态,是老年痴呆的高危人群,研究该病的分类,有助于提前预防病情和延缓痴呆.经研究发现轻度认知障碍和正常人间脑皮层是存在差异的.本文采用多任务学习选择特征,共享任务间的相关性,对脑皮层形态学的三种指标:脑皮层厚度、灰质体积、皮层复杂度看成三个任务,选择特征时任务之间存在一定的联系,与F-score、mRMR两种特征选择方法进行对比.并结合极限学习机和支持向量机分类方法,探究特征选择方法与哪种分类器结合最优.实验证明,多任务学习方法比另外两种特征选择方法效果好,并且和极限学习机分类方法的结合效果最优,在区分轻度认知障碍患者和正常老人上具有更高的准确率和较低的时间复杂度,为实现轻度认知障碍的自动诊断提供了理论依据.

Mild cognitive impairment, which is a middle state between Alzheimer＇s disease and normal aging, is a high risk state of Alzheimer＇s disease. Studying for mild cognitive impairment, will help advance the prevention of illness and delaying dementia. It is found that there is a difference between mild cognitive impairment and normal human in cerebral cortex morphological. In this paper, the feature selection based on multi task learning, It can share correlations between tasks, the three indicators of cerebral cortex mor- phology：cerebral cortex thickness, gray matter volume and the complexity of cerebral cortex as three tasks, which is a certain link be- tween the three tasks in the feature selection. F-score and mRmR, the two feature selection methods are compared with multi task learning. Classification methods used extreme learning machine and support vector machine. We combined three feature selection meth- ods and three classification approaches to determine an optimal combination for this dataset. Results show that multi task learning method perform better than the other two feature selection method, and combined with extreme learning machine classification method is optimal, achieve higher accuracy and lower time complexity in distinguishing mild cognitive impairment patients and normal elderly. The results indicate certain feasibility of the automatic diagnosis of mild cognitive impairment.