Application of nonlinear classifiers with the principal component analysis in fMRI language activation pattern recognition in a multisite study for pediatric epilepsy
Conference
You, X, Guillen, M, Gaillard, WD et al. (2009). Application of nonlinear classifiers with the principal component analysis in fMRI language activation pattern recognition in a multisite study for pediatric epilepsy
. 2 754-758.
You, X, Guillen, M, Gaillard, WD et al. (2009). Application of nonlinear classifiers with the principal component analysis in fMRI language activation pattern recognition in a multisite study for pediatric epilepsy
. 2 754-758.
This paper describes a unique classification approach using nonlinear decision functions (NDF) as means to automatically classify language related functional Magnetic Resonance Imaging (fMRI) brain activation maps into typical and atypical categories involving subjects from a multisite consortium for pediatric epilepsy research. The classification results obtained are compared to the well-established method of support vector machines (SVM), The NDF are evaluated under different degrees of complexity, and assuming any number of dimensions in the decision space, where the leading eigenvectors of the principal component analysis (PCA) are considered. Results reveal high classification accuracy and sensitivity on the testing data in both methods. However, NDF were found to have the best performance with a complexity degree of 7 in a 4D feature space, yielding the following results: 96% accuracy, 97% sensitivity, 95% specificity, and 95% precision. Distinct activation patterns are observed when applying the NDF-based classifier on 122 real datasets.
