基于深度学习与多模态生理数据的阿尔茨海默病分类方法研究  被引量:4

Classification model of Alzheimer's disease based on deep learning and multimodal physiological data

在线阅读下载全文

作  者:王婧萱 王雯婧 闻亮 李贞妮[1] WANG Jing-xuan;WANG Wen-jing;WEN Liang;LI Zhen-ni(College of Information Science and Engineering,Northeastern University,Shenyang 110819,China;Department of Neurosurgery,General Hospital of Northern Theater Command,Shenyang 110016,China;China Medical University,Shenyang 110122,China)

机构地区:[1]东北大学信息科学与工程学院,沈阳110819 [2]北部战区总医院神经外科,沈阳110016 [3]中国医科大学,沈阳110122

出  处:《医疗卫生装备》2023年第11期1-8,共8页Chinese Medical Equipment Journal

基  金:辽宁省博士科研启动基金计划项目(2021-BS-054);中央高校基本科研业务专项资金资助项目(N2204006);辽宁省“兴辽人才计划”项目(XLYC2002109);辽宁省科技民生计划项目(2021JH2/10300059);沈阳市科技计划项目(20-205-4-017)。

摘  要:目的:为了提高阿尔茨海默病(Alzheimer’s disease,AD)的分类效果,提出一种基于深度学习与多模态生理数据的AD分类方法。方法:选用阿尔茨海默病神经影像学计划(the Alzheimer’s Disease Neuroimaging Initiative,ADNI)数据库中AD患者、早期轻度认知障碍(early mild cognitive impairment,EMCI)患者、晚期轻度认知障碍(late mild cognitive impairment,LMCI)患者和正常认知(normal cognition,NC)受试者的多模态数据,利用改进的New_ResNet50网络提取受试者大脑MRI图像特征进行分类,利用3D-Unet-Attention网络对海马体图像进行分割后通过残差网络进行分类,利用多层感知机(multi-layer perceptron,MLP)网络基于患者的生理数据与海马体体积进行分类,并对3个网络给出的分类结果采用投票法确定最终分类结果。比较改进的New_ResNet50网络、3D-Unet-Attention网络分类模型与传统网络分类模型对AD的分类效果,以及融合New_ResNet50网络、3D-Unet-Attention网络、MLP网络的分类模型和单一网络分类模型对AD的分类效果。结果:改进的New_ResNet50网络、3D-Unet-Attention网络分类模型分类准确率较传统网络分类模型均有提高,而融合网络分类模型对AD患者与NC受试者的分类准确率(97.99%)相较New_ResNet50网络、3D-Unet-Attention网络、MLP网络分类模型分别提高了1.51%、1.51%和14.62%。结论:提出的分类方法对AD具有较好的分类效果,可以有效辅助医生诊断。Objective To propose an Alzheimer's disease(AD)classification method based on deep learning and multimodal physiological data.Methods Multimodal data from the Alzheimer's Disease Neuroimaging Initiative(ADNI)database of AD patients,early mild cognitive impairment(EMCI)patients,late mild cognitive impairment(LMCI)patients and normal cognition(NC)subjects were selected.Three networks were used for AD classification,of which an improved New_ResNet50 network extracted the features of MRI images of the subject's brain to realize AD classification,a 3D-Unet-Attention network segmented the hippocampus images and implemented residual network-based AD classification,and a multi-layer perception(MLP)network carried out AD classification based on patient physiological data and hippocampus size,and the final classification results were determined with the voting method.Comparison analyses were performed on the classification results by the improved New_ResNet50 network model,3D-Unet-Attention network model or traditional network models,and the improved New_ResNet50 network model,3D-Unet-Attention network model and MLP network model were all compared with the fusion network model involving in the three networks model above.Results The improved New_ResNet50 network model and 3D-Unet-Attention network model both had the classification accuracy enhanced when compared with the traditional network models,and the fusion network model had a classification accuracy of 97.99%for AD patients and control normal,which was higher by 1.51%,1.51%and 14.62%than those by the improved New_ResNet50 network model,3D-Unet-Attention network model and MLP network model respectively.Conclusion The classification method proposed behaves well for AD classification,and can be used for auxiliary diagnosis of AD.

关 键 词:深度学习 多模态生理数据 阿尔茨海默病 New_ResNet50网络 3D-Unet-Attention网络 MLP网络 

分 类 号:R749.16[医药卫生—神经病学与精神病学] TP18[医药卫生—临床医学]

 

参考文献:

正在载入数据...

 

二级参考文献:

正在载入数据...

 

耦合文献:

正在载入数据...

 

引证文献:

正在载入数据...

 

二级引证文献:

正在载入数据...

 

同被引文献:

正在载入数据...

 

相关期刊文献:

正在载入数据...

相关的主题
相关的作者对象
相关的机构对象