基于脑功能连接和SAE的自闭症分类

Classification of autism spectrum disorder based on brain functional connectivity and SAE

为了对自闭症(autism spectrum disorder,ASD)的计算机辅助诊断提供参考,本文探讨了基于大数据和深度学习的静息态功能磁共振成像(rs-fMRI)的数据分类研究。研究从国际自闭症专业数据库(ABIDE)中获取了306名ASD和350名正常受试者(typically developing,TD)的rs-fMRI数据。通过对预处理之后的rs-fMRI数据提取脑功能连接(Functional connectivity,FC)相关矩阵,再利用堆栈自编码(Stacked autoencoder,SAE)进行训练,最后对ASD和TD进行了分类,得到了准确率高达95.27%的识别。本文的结果表明,基于相关矩阵和SAE的ASD分类系统已经达到了较高性能,可以为计算机辅助诊断ASD提供参考。

In order to provide a guideline to computer-aided diagnosis of autism spectrum disorder(ASD),the techniques of big data and deep learning were utilized to investigate the data classification of the resting state functional magnetic resonance imaging(rs-fMRI).The rs-fMRI data were collected from 306 ASD and350 typically developing(TD)individuals in autism brain imaging data exchange database(ABIDE).The correlation matrix of functional connectivity(FC)was extracted from the preprocessed rs-fMRI data and the training was performed using the stacked autoencoder(SAE).Furthermore,the classification of ASD and TD provided an accuracy of 95.27%.These results showed that the ASD classification system based on the correlation matrix and the SAE has achieved higher performance,and can provide a guidance to the computer-aided diagnosis of ASD.