基于I3D-CNN的自闭症分类方法

Classification of autism based on inflated three dimension convolution neural network

为解决传统基于静态功能网络连接的自闭症分类算法忽略了脑功能连接的时变特性问题,提出一种基于膨胀卷积网络(inflated three dimension convolution neural network,I3D-CNN)的自闭症分类识别方法。提取被试大脑的静息态功能核磁共振影像(rest state functional magnetic resonance imaging,RS-fMRI)每个感兴趣区域(region of interest,ROI)的时间序列,基于时间序列利用随机滑动时间窗口法,构建多个3D动态脑功能连接矩阵,使用I3D-CNN从3D动态脑功能连接矩阵中提取大脑的时空特征,建立自闭症分类模型。通过在ABIDE数据集上进行实验,验证了该方法的可行性和有效性。

To solve the problem that traditional autism classification algorithms based on static functional network connectivity ignore the time-varying characteristics of brain functional connectivity,an autism classification and recognition method based on inflated three dimension convolution neural network(I3D-CNN)was proposed.Time series of each interest(ROI)in tested brain state functional magnetic resonance imaging(RS-fMRI)were extracted.The method of random sliding time window was used based on time series,multiple 3D dynamic brain function connection matrix was constructed,the I3D-CNN was used to extract the characteristics of space and time from 3D dynamic brain function connection matrix,the brain autism classification model was established.Experimental results on the ABIDE data set show the feasibility and effectiveness of the proposed method.