文章摘要

文章信息/Info

Title:: Early behavior recognition method of Alzheimer′s disease based on machine learning

Author(s):: YANG Bangkun1; WANG Lesheng1; NIE Ying2; XIONG Wenping1; 1.Department of Neurosurgery, Zhongnan Hospital of Wuhan University,Wuhan 430071, China;2. Department of Pediatrics,Wuhan No.1 Hospital, Wuhan 430033

Keywords:: Machine learning; Alzheimer′s disease; Incipient behavior; Recognition research; sMRI image; Kernel support vector machine

摘要:: 本研究基于机器学习的阿尔兹海默症初期行为辨识方法，及时发现患者大脑的早期病变，把握最佳治疗机会。从ADNI公共数据库获取阿尔兹海默症、轻度认知障碍以及正常对照组的结构性磁共振成像（sMRI）图像，将其通过Freesurfer软件执行图像平滑、分割、时间层校正等操作，转换为sMRI数据，使用内核局部Fisher判别分析算法提取sMRI数据特征，利用基于核支持向量机的数据分类算法，分类所提取sMRI数据特征，经十折交叉验证评估后，实现阿尔兹海默症初期行为的准确辨识。实验结果表明，该方法的灵敏度、特异性、准确率以及曲线下面积四个指标均保持最高，具有较优异的阿尔兹海默症初期行为辨识效果。该方法不仅能有效辨识患者与健康人，还能正确区分阿尔兹海默症和轻度认知障碍患者，辨识效果显著。

Abstract:: We studied the early behavior recognition method of Alzheimer′s disease based on machine learning, to find the early brain lesions of patients in time, and grasp the best treatment opportunity. The structure magnetic resonance imagin(sMRI) images of Alzheimer′s disease, mild cognitive impairment and normal control group were obtained from the ADNI public database. The images were smoothed, segmented and time corrected by Freesurfer software, and were converted into sMRI data. The kernel local Fisher discriminant analysis algorithm was used to extract the features of sMRI data, and the kernel support vector machine based data classification algorithm was used to classify the extracted sMRI data characteristics, after ten fold cross validation evaluation, the early behavior of Alzheimer′s disease could be accurately identified. The experimental results showed that the sensitivity, specificity, accuracy and area under the curve of the four indicators of this method were the highest, and had a better effect on early behavior recognition of Alzheimer′s disease.This mathod can not only effectively identify patients and normal people, but also correctly distinguish Alzheimer′s disease and mild cognitive impairment patients, and the recognition effect is significant.