基于脑连接网络的阿尔茨海默病临床变量值预测

摘要

Brain functional connectivity networks have been widely used for diagnosing brain diseases.However, a traditional brain network based on classification methods cannot assess the stage or predict the development of the disease.Recent studies show that the values of the clinical variables of brain disease can effectively help doctors evaluate the disease.In this study, a novel brain-connectivity-network-based method was proposed for estimating the values of the clinical variables of Alzheimer's disease.First, the functional connectivity network was extracted from the brain images.Then, LASSO , which is a regression analysis method, was adopted for feature selection and elimination of redundant features;the clustering coefficients and edge weights of the network were fused as features.Finally, support vector machine regression was used to predict the values of the clinical variables.The proposed method was validated on the ADNI dataset, and the experimental results demonstrate that the proposed method can accurately predict the values of clinical variables and verify the effectiveness of the fusion of multiple features.%目前脑功能连接网络已被广泛用于大脑疾病诊断,然而传统的脑网络分类方法无法评估疾病所处的阶段以及预测病情的发展.近期的研究表明,脑疾病的临床变量值可以有效地帮助医生进行疾病评估,为此提出一种基于脑连接网络的方法,用于对阿尔茨海默病临床变量值进行预测.首先从脑影像中提取功能连接网络,然后使用LASSO进行特征选择,剔除不具有判别性的边.同时融合网络的聚类系数和边的权重作为特征.最后使用支持向量回归机预估临床变量值.在ADNI数据集上对提出的方法进行验证,实验结果表明,提出的方法不仅能够准确地预测疾病临床变量值而且还验证了多种特征融合的有效性.