Classification of fMRI Data using Support Vector Machine and Convolutional Neural Network

In recent years convolutional neural network have obtained more popularity because of its progressive performance for different applications especially for object recognition. In neuroimaging, data varies from person to person and condition to condition so it is always a challenging job to model the brain data. Any analysis in neuroimaging is also dependent on the quality of data and currently, functional magnetic resonance imaging is considered as the best among all techniques. It is most reliable and popular modality to measure the brain activity patterns. In fMRI, region of interest is a common method of analysis in which data is taken from a specific brain region based on the structural or functional information. In this study, convolutional neural network is applied to the significant voxels obtained through the t-contrast of the design matrix during the ROI analysis. Data is taken against two conditions and 1000 significant voxels with highest absolute values are taken for each condition for further analysis. During the proposed method, analysis is performed using convolutional neural network along with ROI analysis. Support vector machine is used in the classification of both methods; ROI and proposed methods. In conclusion, it is shown that the features extracted through convolutional neural network can provide better significant results compared to the other one.