文章摘要

文章信息/Info

Title:: Study on factors affecting classification ?of Alzheimer′s disease based on structural MRI

作者:: 黎建忠1; 曾安1; 2; 潘丹3; 4△; Song Xiaowei5; 郭慧6; 王卓薇1; 1.广东工业大学计算机学院,中国广州 510006；2.广东省大数据分析与处理重点实验室,中国广州 510006；3.广东建设职业技术学院现代教育技术中心,中国广州 510440；4.广州市本真网络科技有限公司,中国广州 510095；5.西蒙弗雷泽大学影像技术实验室,加拿大温哥华 V6B 5K3；6. 天津医科大学总医院医学影像科,中国天津 300052

Author(s):: LI Jianzhong1; ZENG An1; 2; PAN Dan3; 4; SONG Xiaowei5; GUO Hui6; WANG Zhuowei1; 1.Faculty of Computer Science, Guangdong University of Technology, Guangzhou 510006, China;2.Guangdong Key Laboratory of Big Data Analysis and Processing, Guangzhou 510006; 3.Modern Education Technical Center, Guangdong Construction Polytechnic, Guangzhou 510440;4.Guangzhou Benzhen Network Technology Co. Ltd., Guangzhou 510095;5.ImageTech Lab, Simon Fraser University, Vancouver V6B 5K3, Canada;6.Department of Medical Imaging, General Hospital of Tianjin Medical University, Tianjin 300052, China

Keywords:: Alzheimer’s disease; Mild cognitive impairment; Structural magnetic resonance imaging; Three-dimensional reconstruction; Support vector machine

摘要:: 本研究提出基于三类解剖特征的SVM建模方法，探索样本、特征及算法选择三个因素，对阿尔茨海默症（AD）及其前驱阶段分类的重要性。该方法以三维重构sMRI后不同大脑区域的灰质体积、皮层表面积及其平均厚度三类特征作为SVM模型的输入参数，并采用十折交叉验证方法对AD患者、轻度认知损害患者和健康者进行分类识别，并与其他文献结果进行比较分析。实验结果表明，为了达到更高的分类准确率，选择合适的样本和特征，比选择算法更重要。此结论为未来AD的计算机辅助诊断研究工作提供了有益的指导。

Abstract:: Aiming at the problem of classifying Alzheimer’s disease (AD) and its prodromal stage, the factors such as training sample selection, feature extraction and classification algorithm selection were studied to exhibit their importance in improving the classification accuracy. Support vector machine (SVM) modeling method based on three types of anatomical features was proposed. Three types of anatomical features (the volume of gray matter, the surface area and the average thickness of the cerebral cortex) in different brain regions were extracted by 3D reconstruction of sMRI images and were utilized to build a SVM model. With the help of 10-fold cross validation, the trained SVM model was employed to classify AD patients, patients with mild cognitive impairment (MCI) and healthy subjects (HC). Compared with other classification results based on different data sets and different features published in other research papers, the experimental results in this study exhibite that it is more important to select the appropriate training data sets and features than to select the classification algorithm. This conclusion might be helpful for the further research on the computer-aided diagnosis of AD.