| 4D卷积神经网络的自闭症功能磁共振图像分类 |
| |
| 引用本文： | 郭磊,王骏,丁维昌,潘祥,邓赵红,施俊,王士同.4D卷积神经网络的自闭症功能磁共振图像分类[J].智能系统学报,2021,16(6):1021-1029. |
| |
| 作者姓名： | 郭磊 王骏 丁维昌 潘祥 邓赵红 施俊 王士同 |
| |
| 作者单位： | 1. 江南大学 人工智能与计算机学院，江苏 无锡 214122;2. 上海大学 通信与信息工程学院，上海 200444 |
| |
| 摘 要： | 静息态功能磁共振图像是随着时间变化的一系列三维图像。已有的3D卷积过程本质上是对三维图像数据或二维图像+时间维数据进行处理，无法有效地融合静息态功能磁共振图像的时间轴信息。为此，本文提出了新型的4D卷积神经网络识别模型。具体而言，通过对输入的fMRI使用四维卷积核执行四维卷积，在自闭症患者的功能磁共振图像中，从空间和时间上提取特征，从而捕获图像在时间序列上的变化信息。所开发的模型从输入图像中生成多个信息通道，最终的特征表示结合了所有通道的信息。实验结果表明，在保证模型泛化性能的前提下，该方法融合了功能像的全局信息，并且采集了功能像随时间变化的趋势信息，进而解决了用卷积神经网络处理三维图像随时间变化的分类问题。
|
| 关 键 词： | 深度学习 卷积神经网络 自闭症 4D卷积 功能磁共振成像 特征提取 特征融合 图像分类 |
Classification of the functional magnetic resonance image of autism based on 4D convolutional neural network |
| |
| GUO Lei,WANG Jun,DING Weichang,PAN Xiang,DENG Zhaohong,SHI Jun,WANG Shitong.Classification of the functional magnetic resonance image of autism based on 4D convolutional neural network[J].CAAL Transactions on Intelligent Systems,2021,16(6):1021-1029. |
| |
| Authors: | GUO Lei WANG Jun DING Weichang PAN Xiang DENG Zhaohong SHI Jun WANG Shitong |
| |
| Affiliation: | 1. School of Artificial Intelligence and Computer, Jiangnan University, Wuxi 214122, China;2. School of Communication and Information Engineering, Shanghai University, Shanghai 200444, China |
| |
| Abstract: | Resting-state functional magnetic resonance images are a series of three-dimensional (3D) images that change over time. The existing 3D convolution processes 3D image data or two-dimensional image and time-dimensional data, but it cannot effectively fuse the time axis information of a resting-state functional magnetic resonance image. To resolve this, a new four-dimensional (4D) convolutional neural network (CNN) recognition model is proposed in this paper. Specifically, by performing a 4D convolution using a 4D convolution kernel on the input functional magnetic resonance imaging, features are spatially and temporally extracted from the functional magnetic resonance image of a patient with autism, thereby capturing information about the changes in the image’s time series. The developed model generates multiple information channels from the input image, and the final feature representation combines information from all channels. The experimental results show that to ensure the generalization performance of the model, the method fuses the global information of the functional image and collects its trend information over time, consequently solving the classification problem of 3D image changes with time using a CNN. |
| |
| Keywords: | deep learning convolutional neural network autism 4D convolution functional magnetic resonance imaging feature extraction feature fusion image classification |
|
| 点击此处可从《智能系统学报》浏览原始摘要信息 |
|
点击此处可从《智能系统学报》下载全文 |
|
