Pattern Recognition And Classification Based On Multi - Channel Near Infrared Spectroscopy Signal Diagnosis

At present, psychiatric disorders such as schizophrenia are largely diagnosed onsymptomatology. Recently pattern recognition approaches to the analysis of neuroimagingdata such as the classification of patients and healthy controls have attracted people’sinterest.Based on near-infrared spectroscopy (NIRS), recent converging evidence has beenobserved that patients with schizophrenia exhibit abnormal functional activities in theprefrontal cortex during a verbal fluency task (VFT). Therefore, some studies haveattempted to employ NIRS measurements to differentiate schizophrenia patients fromhealthy controls with different classification methods. However, these methods wereconducted on different data, which made it difficult to compare their respectiveclassification performances.The hemoglobin response was measured in the prefrontal cortex of schizophreniapatients and healthy controls during the Chinese version verbal fluency task using amultichannel NIRS system and recruited a large sample of120schizophrenia patients and120healthy controls. The brain activations were investigated during the task period withinschizophrenia group and healthy group, respectively, and compared their relative changesbetween two groups by using the method of statistical analysis. The results confirmed thatpatients with schizophrenia had the significant lower brain activations in the prefrontalcortex and superior temporal cortex.In this study, the classification performance of four classification methods (includinglinear discriminant analysis, k-nearest neighbors, Gaussian process classifier, and supportvector machines) on an NIRS aided schizophrenia diagnosis based on the above findingswere evaluated. Features for classification were extracted from three types of NIRS data ineach channel. A principal component analysis (PCA) were performed for feature extractionand selection prior to comparison of the different classification methods subsequently.A maximumaccuracy of85.83%and an overall mean accuracy of83.37%wereachieved using a PCA-based feature selection on oxygenated hemoglobin signals andsupport vector machine classifier. Comparison with existing methods, this is the firstcomprehensive evaluation of different classification methods for the diagnosis ofschizophrenia based on different types of NIRS signals.Our results suggested that, using the appropriate classification method, NIRS has the potential capacity to be an effectiveobjective biomarker for the diagnosis of schizophrenia.