Brain visual state classification of fMRI data using fuzzy support vector machine

Abstract

The fMRI (Functional Magnetic Resonance Imaging) technology is a revolutionary tool that has lit up the studies of human cognitive processing with the help of efficient methods of image and data analysis. Machine learning classifiers are widely employed to extract all sorts of information from neuroimaging data. This study aims to identify tangible patterns in the fMRI data for visual activity and perform multivariate pattern analysis. It is done by selecting relevant features to indicate the response to visual stimulus of a set of objects belonging to eight different categories. The task intends to identify the nature of the response to the stimuli and classify them according to the brain’s neural activation to the visual stimuli. An SVM (Support Vector Machine) classifier and an FSVM (Fuzzy Support Vector Machine) classifier are implemented to perform the classification based on the features. The training of the classifiers involved 72 test samples per category. The 24 test samples of each category were tested with each of the classifiers. Conclusively, for this dataset, the FSVM classifier performs better than SVM classifier with an increased accuracy of 4% and classifying certain categories with improvement.