文章摘要

文章信息/Info

Title:: Early diagnosis of Alzheimer′s disease based on convolutional recurrent neural network

作者:: 曾安1; 2; 黄殷1; 潘丹3; 4△; SONG Xiaowei5; 1.广东工业大学计算机学院，广州 510006；2.广东省大数据分析与处理重点实验室，广州510006；3.广东建设职业技术学院现代教育技术中心，广州510440；4.广州市大智网络科技有限公司，广州 510000；5.西蒙弗雷泽大学影像技术实验室，温哥华V6B 5K3

Author(s):: ZENG An1; 2; HUANG Yin1; PAN Dan3; 4; SONG Xiaowei5; 1. Faculty of Computer, Guangdong University of Technology, Guangzhou 510006, China；2.Guangdong Key Laboratory of Big Data Analysis and Processing, Guangzhou 510006；3.Modern Education Technology Center, Guangdong Construction Polytechnic, Guangzhou 510440；4.Guangzhou Dazhi Networks Technology Co. Ltd., Guangzhou 510000；5.ImageTech Lab, Simon Fraser University, Vancouver V6B 5K3, Canada

Keywords:: Alzheimer′s disease; Convolutional neural network; Recurrent neural network; Magnetic resonance imaging; Positron emission tomography; Image classification

摘要:: 早期准确诊断能延迟阿尔茨海默症（Alzheimer′s disease，AD）病情的恶化。磁共振成像（MRI）和正电子发射断层扫描（PET）已被证明有助于了解AD相关的解剖和功能性神经变化。近期研究表明，多模态特征的融合可以提高分类性能。本研究提出了一种基于卷积循环神经网络的多模态数据分类新框架，新框架结合了2D卷积神经网络和循环神经网络，以学习3D MRI和3D PET图像切分为2D切片序列之后的切片内、切片间特征，完成AD的早期诊断。本研究方法在AD与NC的分类实验中ACC为93.3%，AUC为98.1%；在MCIc与NC的分类实验准确率为83.8%，AUC为91.9%；MCIc与MCInc的分类实验准确率为79.0%，AUC为88.9%。结果表明该方法具有良好的分类性能。

Abstract:: Early and accurate diagnosis can delay the deterioration of the Alzheimer′s disease. Structural and functional neuroimaging images such as magnetic resonance imaging (MRI) and positron emission tomography (PET) have been shown to be useful in understanding AD-related anatomical and functional neurological changes. We proposed a new framework for multi-modal data classification based on convolutional recurrent neural networks. It combined 2D convolutional neural network (CNN) and recurrent neural network (RNN) to learn 3D MRI and 3D PET images into 2D slice sequences and slice features, to complete early diagnosis of AD. In the classification experiment of AD and NC, the ACC and AUC of this method were 93.3%， 98.1%。The accuracy of the classification experiments in MCIc and NC was 83.8%，the AUC was 91.9%. The accuracy of the classification experiments in MCIc and MCInc was 79.0%，the AUC was 88.9%. The results show that this method has good classification performance.