Multi-classification Pattern Analysis Of Functional Magnetic Resonance Imaging

Resting-state functional Magnetic Resonance Imaging (rs-fMRI) data containsplentiful neural information, and it has been one of the most effective techniques onbrain research. In recent years, pattern recognition methods have been widely used inrs-fMRI study.rs-fMRI data can be obtained without complex experimental design and operation.And it does not require participants to perform any complex cognitive tasks during thescanning. rs-fMRI has been widely used in the brain research. However, the rs-fMRIdata were a combination of neural activity signals, magnetic saturation effects andphysiological noise. The signals we are interested are mixed with the noise, and evenmay be overwhelmed by the noise. The signal-to-noise ratio of rs-fMRI data is very low.Therefore, extracting effective information from rs-fMRI data has always been one ofthe most difficult problems needed to be solved in rs-fMRI study.Multivariate pattern analysis based on machine learning has aroused great interestfor its capacity of finding valuable neuroimaging-based biomarkers. Previous studiesalmost focused on two classification problems. With the deep study, multiclass problemof rs-fMRI has gradually come forth. On the basis of rs-fMRI, the current study aims atinvestigating the multi-classification problem of rs-fMRI by using pattern recognitionmethods.In the current study, principal component analysis (PCA) along with linear supportvector machine (SVM) was used to solve the multi-classification problem among theschizophrenia, their healthy siblings, and the healthy controls. Our classification resultssuggested that the multiclass pattern analysis methods reliably capture discriminativeresting-state functional connectivity patterns among the three groups. And based on theclassification results, we further identified three types of functional connectivity-basedsignatures: i) relating to the state of having schizophrenia, ii) reflecting the geneticvulunerability to develop schizophrenia, and iii) underlying special brain connectivitiesby which healthy siblings overcome genetic risk for developing schizophrenia. Ourinvestigation suggested that the healthy siblings had potential higher risk to developschizophrenia than the healthy controls do, and that the brain connectivity patterns ofthe three groups are different.To our knowledge, the current study was the first to demonstrate the feasibility ofdiscriminating major depressive disorder patients from schizophrenic patients usingwhole-brain resting-state functional network. Especially, this study was the first usageof IDA method in discriminative analysis of rs-fMRI data. And the performance of IDAalgorithm was compared with that of PCA. Based on IDA method，We designed adata-driven multiclass classifier and successfully extracted the significant discriminative functional connections underlying spontaneous neural activity in the brains of the majordepressive disorders and the schizophrenia. We hoped that this new attempt may bringsome practical ideas in dealing with the problem of dimensionality reduction of fMRIdata. And we further found that the major depressive disorders and the schizophrenicpatients both showed common altered functional connections within or across the DMNand cerebellum compared with the healthy controls, which might shed new light on thecommon pathological mechanism underlying the similar clinical behaviors of the twodisorders. These results also gave further evidence supporting the viewpoint that theresting-state functional network is potentially useful in revealing the pathophysiology ofmajor depressive disorder and schizophrenia. In addition, the affective network played anotable role in distinguishing the depression from schizophrenia, perhaps providingpotential biomarkers in the diagnosis between them.Through multi-class discriminative analysis, we have deepened our comprehensionon the multi-classification problem of fMRI data. In addition, we investigate thepathophysiological mechanism and the genetic risk of schizophrenia based onwhole-brain rs-fMRI. We also investigated the common and different pathologicalmechanism underlying schizophrenia and major depressive disorder. Our study played adistinct role in the multi-class pattern analysis of rs-fMRI.