Multivariate Biomedical Data Classification Research Based On Machine Learning Methods

With the development of natural sciences and the advancement of global informatization,the biomedical field has entered the era of big data.Especially after the emergence of machine learning methods,its powerful feature extraction ability has greatly promoted the research process of biomedical data.According to the number of different variables studied,biomedical data can be divided into univariate biomedical data and multivariate biomedical data.Among the various types of multivariate biomedical data,the study of brain-computer interface(BCI)technology helps people with impaired audiovisual perception and limited limb movement to return to normal,and functional Magnetic Resonance Imaging(f MRI)studies the diagnosis and prevention of brain diseases.Therefore,the analysis and processing of these two types of data has received more and more attention from researchers in recent years.Based on machine learning theory,this paper proposes a new method to classify steadystate Visually Evoked Potentials(SSVEP)data in BCI and Autism Brain Imaging Data Exchange(ABIDE)data in f MRI,which effectively improves the classification performance.The main work of this article includes the following two aspects:(1)A classification algorithm based on kernel matrix is proposed for SSVEP data.The existing classification methods usually use the covariance matrix to extract features from SSVEP data,and then use classical machine learning methods to classify.We note that traditional covariance matrices are only a special form of kernel matrices,and that Riemann spaces are more reflective of intrinsic distances than traditional Euclidean spaces,but most classical classification algorithms are based on Euclidean spaces.Therefore,this paper proposes a method for classifying kernel matrices in Riemannian tangent space.First,the Gaussian kernel matrix is used instead of the traditional covariance matrix for feature extraction of SSVEP data;then,the result of feature extraction is mapped into the tangent space of the Riemann manifold;and finally,classification is implemented using a classical machine learning classifier.Experimental results show that the kernel matrix classification algorithm based on Riemannian cut space can effectively improve the classification accuracy.(2)A classification model based on long short-term memory network(LSTM)is proposed for ABIDE data.At present,the classification of ABIDE data is mainly based on feature extraction through different functional connection methods,and then models such as Support Vector Machines(SVM)are constructed to classify them.However,f MRI data is essentially a multivariate time series,and traditional machine learning models may not be able to fully tap into its implicit time series characteristics,while LSTM networks can capture and learn data characteristics from time series data well and take into account time dependence.Therefore,this paper proposes an ABIDE data classification model based on LSTM network.First,the ABIDE data is masked using the Smith2009 70 RSNs atlas;then,a network model containing three layers of LSTM is constructed;and finally,its feasibility is verified in the classification task.Experimental results show that the average accuracy of classification of the model has been improved,but its sensitivity and specificity need to be improved.