Multivariate Pattern Classification Analysis Of Human Brain MRI Data

Magnetic resonance imaging (MRI) is an important brain imaging method in neuroimaging, which has been widely used in cognitive neuroscience and brain disease research. The traditional approaches to the analysis of functional and structural magnetic resonance imaging data usually carry out hypothesis test on each voxel in the brains separately. However, the distributed spatial patterns of these MRI images may be a more appropriate metric for discriminating different task conditions and different groups of subjects. Therefore multivariate pattern classification analysis (MVPA) and other machine learning algorithms are used to identify the spatial patterns in the human brain MRI images to decode the brain states, or distinguish between healthy and disorder brain. This paper studies the application of multivariate pattern classification analysis of three categories of MRI data:decoding individual finger movements based on functional magnetic resonance imaging (fMRI); automatic classification of healthy children and children with primary nocturnal enuresis (PNE) based on resting-state functional magnetic resonance imaging (resting-state fMRI); automatic classification of healthy children and children with attention-deficit hyperactivity disorder (ADHD) based on diffusion tensor imaging (DTI).MVPA-based fMRI can be used to implement the brain-computer interaction, which may decode limb movement information in a non-invasive way from the distributed spatial patterns of brain activation to control computer or external device. The fingers are arguably the most important body parts for fine control and action; therefore we study the decoding of the individual finger movements from fMRI single-trial data. Thirteen healthy human subjects participated in a visually cued delayed finger movement task, and only one slight button press was performed in each trial. The region of interest based MVPA was used to classify the brain fMRI signal patterns excited when five fingers were moved individually, and the decoding accuracy (DA) was calculated for each brain region of interest separately. In addition, the searchlight multivariate pattern classification analysis was employed to search for informative regions in an unbiased manner across the whole brain, and the DA map was calculated for each subject. The results showed that individual finger movements can be decoded based on fMRI single trial during which a single press was performed. We obtained a63.1%average DA (84.7%for the best subject) for the contralateral primary somatosensory cortex (SI) and a46.0%average DA (71.0%for the best subject) for the contralateral primary motor cortex (M1). The permutation test showed that decoding accuracy obtained in these areas is significantly higher than the level of chance decoding (20%). Although the decoding accuracy was not accurate enough to realize a practical BCI system, the results suggest that the non-invasive fMRI technique may provide informative features for decoding individual finger movements and the potential of developing an fMRI-based BMI for finger movement. Beyond decoding this study will allow a better understanding the basis neural mechanism controlling the movements of individual fingers.In the resting-state fMRI based classification of healthy children and children with PNE, the functional connectivity (FC) coefficient between any two of the90brain regions of the AAL atlas was calculated. The feature vectors were constructed with the functional connectivity among all brain regions for each subject, which were used in the multivariate pattern classification between the two types of subjects. The classification accuracy obtained is:80%for the healthy children;70%for the children with PNE. The permutation test showed that the classification accuracy is significantly higher than the level of chance classification (50%). The results imply that the patterns of functional connectivity among all regions of the brain under the resting state may capture the feature of the disease of PNE. Our findings and the proposed method might contribute to development of the MRI data based medical diagnosis. In the DTI based classification of healthy children and children with ADHD, the FA skeleton in the TBSS method was used to extract DTI feature vector for each subject, and the multivariate pattern classification analysis was carried out for five kinds of DTI-derived brain images for three pairs of contrasting groups. We found that the best classification results are obtained when using the SVM classifier with the quadratic kernel function to classify the FA feature vectors of healthy children and children with combination subtype of ADHD, and the classification accuracy obtained are:73.33%for the healthy children;60%for the children with combination subtype of ADHD. However, the permutation test showed that the classification accuracy for children with combination subtype of ADHD is not significantly higher than the level of chance classification (50%). For clinical utility, the obtained classification accuracy is still not accurate enough. We analyzed the causes of the low rate of correct classification, and proposed some methods for improvement. Our findings and the proposed method might contribute to the understanding of the causes of ADHD and development of the MRI data based medical diagnosis.