脑部PET图像在阿尔茨海默病早期诊断中的应用

目的本研究使用脑部正电子发射型计算机断层显像(positron emission computed tomography,PET),并且设计了一个3D卷积神经网络(convolutional neural networks,CNN),以实现对阿尔茨海默病(Alzheimer disease,AD)的早期诊断。方法研究数据取自美国国立卫生研究院老年研究所的ADNI(Alzheimer’s Disease Neuroimaging Initiative)数据库,PET图像和磁共振(magnetic resonance,MR)图像均有收集并对数据进行相关预处理。为避免过早的下采样给模型性能带来不利影响,设计了一个3D CNN模型,比较两种不同模态的数据在AD早期诊断中各自的优缺点。结果使用本研究组设计的3D CNN模型在基于PET图像的AD早期诊断实验中,预测准确率、灵敏度、特异度以及曲线下面积(area under curve,AUC)分别达到71.19%、79.29%、61.35%、71.09%。此外,对本研究组的模型与计算机视觉中的经典模型VGG和ResNet使用相同数据进行对比实验,许多评价指标都要更优。结论使用脑部PET图像并结合3D CNN可以更好地利用3D图像的空间位置信息,更有效地提取特征,能对AD早期的病变情况有更准确高效的识别。

Application of brain PET images in the early diagnosis of Alzheimer disease

Objective In this study,we used positron emission computed tomography(PET)of the brain and designed a 3D convolutional neural networks(CNN)to achieve early diagnosis of Alzheimer disease(AD).Methods Data were obtained from the Alzheimer’s Disease Neuroimaging Initiative(ADNI)database.Both PET images and magnetic resonance(MR)images were collected and preprocessed.In order to avoid premature downsampling from affecting the model performance,we designed a 3D CNN model to compare the advantages and disadvantages of the two different modal data in the early diagnosis of AD.Results By using our designed 3D CNN model in the early diagnosis experiment of AD based on PET images,the prediction accuracy,sensitivity,specificity and the area under curve(AUC)reached 71.19%,79.29%,61.35%and 71.09%,respectively.Compared with VGG and ResNet,our model was better than these two models in many evaluation indicators.Conclusions Combining brain PET images and 3D CNN can make better use of the spatial position information of 3D images,extract features more effectively,and can more accurately and efficiently identify the pathological changes in the early stage of AD,which is helpful for timely detection of diseases.