【作者】 王彦；

【导师】 马婷；

【作者基本信息】 哈尔滨工业大学 ， 信息与通信工程， 2019， 硕士

【摘要】 随着世界人口老龄化,阿尔茨海默病(Alzheimer Disease,AD)成为老年人生命健康的重大威胁。目前AD诊断的方法主要依靠心理学量表和医生的经验进行判断,诊断准确率不高。多模态磁共振成像技术的出现给患者提供了一种无创安全的脑疾病探查手段。其中,弥散张量成像(Diffusion Tensor Imaging,DTI)可以反映脑神经纤维的密度与分布,功能磁共振成像(functional Magnetic Resonance Imaging,fMRI)则反映了人脑神经活动情况。同时,AD的产生与脑网络异常密切相关,但目前针对脑网络的研究较少。本文将利用多模态磁共振成像数据完成脑网络的建模与探究,通过寻找正常人与AD患者脑网络之间存在的差异为AD诊断提供帮助,解决AD诊断准确率不高的问题。本文完成了脑网络的建模并对脑网络进行了探究。根据fMRI数据和DTI数据不同的物理意义,本文分别建立了脑功能连接网络和脑白质结构连接网络。脑功能连接网络可以反映脑功能之间的同步性,脑白质结构连接网络则可以反映脑区间实际存在的纤维连接情况。通过建立脑网络,本文将磁共振成像与脑网络研究联系起来。本文用图论的方法对构建的脑网络进行分析,分别提取了两种模态脑网络的节点度、节点效率、节点介数中心性三类参数。利用T检验的方法对正常人与AD患者的图论参数的差异性进行分析,将图论参数具有显著性差异的脑区挑出来作为脑网络特征。此外,本文利用支持向量机对挑选出来脑网络特征进行了合理性验证。将挑选出的两种脑网络特征作为标准,选用K折交叉验证的方法对样本数据进行训练与分类测试,两类样本均取得了较高的分类准确率。根据上述方法,本文提出了一种构建AD辅助诊断模型的设想。本文提出了一种利用深度学习对脑网络进行分析的方法。该方法可以通过自主学习找寻特征,同时也可以将多模态的脑网络进行联合探究。本文根据处理数据的特点构建了一个新的卷积神经网络框架用于分析脑网络。同时本文将脑功能连接网络与脑白质结构网络作为输入对框架进行训练及测试。通过测试集的验证,用该框架进行网络分析有较高的准确率。框架在多模态脑网络联合作为输入时准确率最高,比单一网络作为输入时效果好。根据该方法,本文提出了另一种构建AD辅助诊断模型的设想。更多还原

【Abstract】 With the aging of the world’s population,Alzheimer’s disease(AD)has become a major threat to the elderly.At present,AD diagnosis mainly depends on psychology scales and subjective judgment of doctors,the diagnostic accuracy is insufficient.MRI provides us with a non-invasive brain disease examination.DTI can reflect the density and distribution of nerve fibers in the brain,while fMRI can reflect the brain activity of human brain.At the same time,the generation of AD is closely related to the abnormality of the brain network.However,there are currently few studies on brain networks.This paper can use multimodal MRI data to construct and explore brain networks,then find out the differences in brain networks between normal people and AD patients,and use this result to improve the diagnostic accuracy of AD.This paper completed the modeling of brain network and explored the brain network.According to the different MRI data,this paper constructed the functional connectivity network(FCN)and the DTI structural connectivity network(DTISCN).The FCN can reflect the synchrony between brain functions,and the DTISCN can reflect the actual fiber connection in the brain region.This paper linked MRI with brain network research by constructing brain network models.This paper analyzed the constructed brain network by graphical method,and extracted the parameters of nodal degree,nodal efficiency and nodal betweenness centrality of each brain region.The T-test method was used to analyze the graphical parameters of normal people and AD patients,and corresponding parameters with significant differences were used as brain network features.In addition,the support vector machine was used to verify the rationality of the selected brain network features.The two types of brain network features were selected as the standard and the K-fold cross validation method was used to train and classify the sample data.Both types of samples achieved high classification accuracy.This paper proposed the idea of constructing an AD-assisted diagnostic model based on this method.This paper proposed a method for analyzing brain networks using deep learning.This method can find features by autonomous learning,and can perform joint analysis on FCN and DTISCN.This paper constructed a new convolutional neural network model for analyzing brain networks based on the characteristics of processing data.We used the FCN and DTISCN as input to train and test the model.Through the verification of the test set,the network analysis with the model has a high accuracy.The model has the highest accuracy when combined with multimodal brain networks as input,and works better than a single network as input.This paper proposed another idea for constructing an AD-assisted diagnostic model based on this method.更多还原