Diagnosis of Schizophrenia Based on Deep Learning Using fMRI

Abstract

Schizophrenia is a brain disease that frequently occurs in young people. Early diagnosis and treatment can reduce family burdens and reduce social costs. There is no objective evaluation index for schizophrenia. In order to improve the classification effect of traditional classification methods on magnetic resonance data, a method of classification of functional magnetic resonance imaging data is proposed in conjunction with the convolutional neural network algorithm. We take functional magnetic resonance imaging (fMRI) data for schizophrenia as an example, to extract effective time series from preprocessed fMRI data, and perform correlation analysis on regions of interest, using transfer learning and VGG16 net, and the functional connection between schizophrenia and healthy controls is classified. Experimental results show that the classification accuracy of fMRI based on VGG16 is up to 84.3%. On the one hand, it can improve the early diagnosis of schizophrenia, and on the other hand, it can solve the classification problem of small samples and high-dimensional data and effectively improve the generalization ability of deep learning models.

Figure 1

Convolutional neural network structure.

Figure 2

VGG16 structure.

Figure 3

Feature visualization.

Figure 4

The improved VGG16 schizophrenia classification model.

Figure 5

Transfer learning training process.

Figure 6

Data preprocessing process.

Figure 7

The ROC curves of the four models.