基于多输出的3D卷积神经网络诊断阿尔兹海默病

Diagnosis of Alzheimer's Disease Based on Multi-Output Three-Dimensional Convolutional Neural Network

随着人口老龄化的加深,阿尔兹海默疾病更加大众化地出现在我们生活中,而早期精准诊断阿尔兹海默疾病并进行正向干预可有效延缓阿尔兹海默疾病的进程.基于磁共振图像的阿尔兹海默疾病的精准诊断需要综合利用多个感兴趣区域(ROIs)的信息,而单个ROI无法体现不同ROIs之间存在的联系与影响.本文首先提出三输入3D卷积神经网络(CNN),综合利用大脑3D磁共振图像中海马体、灰质(无海马体)和白质3个ROIs的信息.此外,随着神经网络的加深,原始图像的重要特征信息会部分丢失,因此我们又提出一种多输出3D CNN,通过增加中间层的连接和输出,缩短输入和输出之间的距离,增强特征传播,减少特征信息的丢失.结果显示采用多输出3DCNN模型实现整个测试集三分类的准确率为90.5%、精确率为91.0%、灵敏度为90.4%、特异性为95.2%、F1-score为90.5%,诊断性能优于单输出3D CNN模型.

Alzheimer's disease has become more prevalent in our lives as the population ages.Accurate diagnosis and positive intervention can effectively delay the progress of early-stage Alzheimer’s disease.Accurate diagnosis of Alzheimer’s disease requires the combination of information from multiple regions of interest(ROIs),because the use of one single ROI may lose the connection and impact among multiple brain regions.In this paper,we firstly proposed a three-input convolutional neural network(CNN)to comprehensively utilize the information from three ROIs,hippocampus,other gray matter(without hippocampus)and white matter.In addition,as the neural network deepens,important feature information of original image will be partially lost.Therefore,we proposed a multi-output 3D CNN,which increases the connection and output of middle layers,shortens the distance between input and output,enhances feature propagation and reduces the loss of feature information.The results showed that the accuracy rate,precision rate,sensitivity,specificity and F1-score of the test set diagnosis obtained by multi-output 3D CNN model were 90.5%,91.0%,90.4%,95.2%and 90.5%,respectively.The diagnostic performance was better than that of the single-output 3D CNN model.