Machine Learning Research Based On Human Brain Magnetic Resonance Imaging

The human brain,as one of the most complex systems in the world,has more than hundreds of billions connection.The brain is the source of human emotions,thinking,cognition,behavior,emotions,etc.It is the last code book left by nature for humans.magnetic resonance imaging(MRI)technology has laid a solid technical foundation for us to uncover the mysteries of the human brain.For example,structural magnetic resonance imaging(sMRI)can be used to reflect the characteristics and changes in the structure of the human brain,while functional magnetic resonance imaging(fMRI)can reflect the functional activity pattern of the human brain.To help people better understand the human brain's thinking patterns and activity rules.In recent years,the rise of machine learning methods has attracted widespread attention in the field of computer vision and pattern recognition.Compared with the traditional statistical analysis methods,machine learning methods can extract more invisible information from the human brain,providing a new perspective and platform for pattern classification and differences analysis between groups.In this paper,three machine learning methods are used,namely three-dimensional feature extraction technology,group-wise sparse coding technology,and deep learning technology to solve gender discrimination in human brain,analysis of differences between schizophrenia groups,and multi-center big data brain image classification problems.The paper mainly includes the following three aspects of the work.3D Feature Extraction for Gender Identification Based Human Brain MRI.Determining gender by examining the human brain is not a simple task because the spatial structure of the human brain is complex,and no obvious differences can be seen by the naked eyes.In this paper,we propose a novel three-dimensional feature descriptor,the three-dimensional weighted histogram of gradient orientation(3D WHGO)to describe this complex spatial structure.The descriptor combines local information for signal intensity and global three-dimensional spatial information for the whole brain.We also improve a framework to address the classification of three-dimensional images based on MRI.This framework,three-dimensional spatial pyramid,uses additional information regarding the spatial relationship between features.The proposed method can be used to distinguish gender at the individual level.We examine our method by using the gender identification of individual magnetic resonance imaging(MRI)scans of a large sample of healthy adults across four research sites,resulting in up to individual-level accuracies under the optimized parameters for distinguishing between females and males.Compared with previous methods,the proposed method obtains higher accuracy,which suggests that this technology has higher discriminative power.With its improved performance in gender identification,the proposed method may have the potential to inform clinical practice and aid in research on neurological and psychiatric disorders.Research on schizophrenia based on sparse coding.Resting state functional magnetic resonance imaging data has become a powerful technique for analyzing cognitive and psychiatric disorders,including schizophrenia.The past group analysis methods are based on the research of brain regions in a certain brain function network in the brain.However,recent studies have shown that some brain regions participate in the activities of multiple functional networks at the same time.In this paper,we propose a new method of machine learning,called sparse coding between groups,to find differences between schizophrenia patients and healthy controls in resting state fMRI signals.Our analysis is based on the sparse expression of the whole brain voxel signal at the group level.This chapter studies the application of sparse coding methods in the analysis of schizophrenia,and seeks to find traits that distinguish schizophrenia from normal people through such methods.First,we extracted the fMRI signals from all the subjects.These data were mapped to the standard MNI template space through the preprocessing process to construct a large-scale input signal matrix.Second,we use dictionary learning and sparse coding to obtain a matrix of coefficients.Next,we used a two-sample t-test to analyze the pattern of increased and decreased activity in schizophrenic patients compared to normal subjects.Finally,the AAL template was used to analyze the distribution of the detected area.We tested our method on the COBRE data set and the experimental results showed that the schizophrenia patients showed an enhanced pattern in the nominal attention network compared to normal subjects.In addition to the nominal network,others show a weakening pattern.These results also provide a new perspective for a better understanding of schizophrenia.Multi-center Brain Neuroimaging Classification Based on Deep Learning.With the development of neuroimaging technology,more and more magnetic resource imaging(MRI)data are acquired.Traditional computational analysis methods based on single site and small samples are facing great challenges.Deep learning technology,which is born with artificial intelligence,has shown powerful ability to solve classification problem based on big data in many researches,while not been widely used in neuroimaging classification.In this paper,we propose a novel 3D deep adding neural network(3D DANet)to classify samples' gender label on 6008 MRI samples from over 61 sites in six datasets.The proposed method utilizes multiple convolutional layers to extract the gradient information in different orientations and combines spatial information in two scales through adding operation.High accuracy(over 92.5%)is obtained to test the classification performance using gender label with a standard 5-fold cross validation strategy,which means that the proposed method can effectively handle big data classification on multi-center.Compared with some traditional classification methods and some deep learning architectures,the proposed method obtains higher accuracy,suggesting its stronger power to distinguish gender.The results of leave-one-subset out validation prove the transfer ability of the proposed method.To the best of our knowledge,our work is the first to classify gender on such a large-scale data from multiple centers with such a high accuracy.With its improved performance in classification and its transferable program codes,the proposed method may have the potential to be used in intelligent medical treatment and clinical practice based on mobile terminal.