Automatic Schizophrenia Discrimination Based On Functional Near-infrared Spectroscopy

Schizophrenia is a kind of serious mental illness,not only causes a lot of pain for patients but also brings a heavy burden to family and society.Without objective physiological data supporting,the traditional subjective diagnosis way can easily lead to misdiagnosis.Functional near-infrared spectroscopy(fNIRS)has been widely used in clinical diagnosis for its advantages such as non-invasive,portable,real-time and low cost.This paper completes the study of automatic identification the schizophrenia patients on the basis of fNIRS data,which has an important reference significance to the clinical diagnosis.In this paper,the GLM feature extraction and extraction method based on analysis brain networks attributes are used to analyze the fNIRS data feature,the SVM is used to identify the schizophrenic.The major research work are as follows:(1)fNIRS data preprocessing.In order to eliminate baseline drift and high frequency noise,the original data are analyzed by Discrete Fourier Transform to get the frequency range.According to the frequency range,a low-pass filter is designed to filter the signal component which is outside of the frequency range.After this,the signal to noise ratio is improved.(2)Feature extraction and classification based on general linear model(GLM).Using GLM to extract the value of 52 channel activation and then testing the 52 channel of the patients group and the normal control group by independent sample T test.The obvious difference channels are selected from the 52 channels to construct feature vector.The leave-one-out cross validation method is used to test the SVM classifier by constructing the sample library(including 34 cases of normal control and 42 cases of schizophrenia patients),the recognition accuracy rate reaches 88.15%,specificity reaches 100%,and sensitivity reaches 78.57%.(3)Feature extraction and classification based on brain networks.The normal people and schizophrenia patients’ brain networks are constructed in this paper.And then the attributes of each brain networks are computed,which include nodal degree,clustering coefficient,local efficiency and global efficiency.Through the comparative analysis of a single attribute,nodal degree which is the most suitable classification attribute is selected to construct the feature vector for classification.And the combined feature of multiple attributes is also experimented.Finally,the nodal degree obtains better classification effect,recognition accuracy is 85.53%,specificity is 76.47%,sensitivity is 92.86%.The method proposed in this paper has implemented the automatic discrimination of patients with schizophrenia.Through the experiments on the data set,this research has reached the expected goals.It could provide an objective data for the diagnosis of schizophrenia and help the doctor improve the diagnostic accuracy.