Research On Classification Method Of FMRI Data Based On Broad Learning System

The fMRI data classification refers to a high-efficiency medical technology method of identifying the subject’s functional magnetic resonance imaging(fMRI)data by computer means to automatically diagnose whether the subject has a corresponding brain disease.This technology provides an important means to crack the pathogenesis of related brain diseases,and at the same time can provide early screening and diagnosis for people at high risk of disease,so it has great research value and application prospects.In recent years,a large number of researchers have applied machine learning methods to fMRI data classification,and attracted widespread attention.These methods can be roughly divided into traditional machine learning-based and deep learning-based fMRI data classification methods,in which deep learning-based fMRI data classification methods can effectively extract deep-level features in fMRI data,and are better than traditional machine learning methods.The classification performance is gradually favored by researchers and has become a hot spot in the field of fMRI data classification.Although the deep learning method effectively makes up for the lack of model’s ability to fit the high-dimensional small sample characteristics of fMRI data in the traditional machine learning method,the problems such as slow training speed and long time are caused by the generally complex training structure and a large number of computational parameters.In order to overcome the above shortcomings,this paper has completed the following two aspects of work:(1)In order to improve the speed of fMRI data classification,an fMRI data classification method based on Single Broad Learning System(SBLS)is proposed.In this method,a new kind of machine learning method,Broad Learning System(BLS),is adopted.BLS can extract the deep features of fMRI data through simple structure and accelerate the classification speed.To be specific,firstly,the data was constructed by using the voxel of the region of interest of fMRI.Then,through random feature mapping and feature enhancement,the shallow and deep features in the data are extracted respectively,and the model framework is constructed.Finally,the ridge regression inverse is used to calculate the connection weight of the classification model to realize the classification of the fMRI data.In this paper,three data sets of ABIDE I,ABIDE II and ADHD-200 are used to compare the proposed method withsix classification methods.The results show that the method proposed in this part can greatly reduce the training time while maintaining good classification accuracy.(2)In order to further improve the classification performance of the fMRI data classification method based on Single Broad Learning System,we propose the fMRI data classification method based on Multilevel Broad Learning System(MBLS).MBLS realizes a multi-level BLS model that can fit the complex connections of human brain by constructing the BLS in series within layers and in parallel between layers,aiming to ensure efficient classification speed and enhance the extraction ability of connectivity features in data.Firstly,the Pearson correlation coefficient of fMRI are used to construct the data.Then,the connectivity features in each brain region and between different regions in the data are extracted through hierarchical feature mapping and feature enhancement,and the model framework is constructed.Finally,the ridge regression inverse is used to calculate the connection weight of the classification model to realize the multi-layer width model structure.Experimental results on ABIDE I and ABIDE II data sets show that,compared with SBLS method,MBLS method can effectively improve the performance of fMRI data classification.By comparing the proposed method with six classical classification methods,the MBLS method can greatly reduce the training time while improving the classification accuracy.The work which carried out in this subject not only extends the fMRI data classification method to the field of Broad Learning,promotes the research and development of Broad Learning in the fMRI data classification,but also plays a positive role in the rapid and accurate diagnosis of brain diseases.