Feature selection and classification of aMCI subjects using local fMRI activation patterns
Mingwu Jin1, Xiaowei Zhuang2, Tim Curran3, and Dietmar Cordes2
1University of Texas at Arlington, Arlington, TX, United States, 2Cleveland Clinic Lou Ruvo Center for Brain Health, Las Vegas, NV, United States, 3University of Colorado Boulder, Boulder, CO, United States
Two feature selection
methods and four classification methods were applied to fMRI memory activation
data obtained from two groups of amnestic MCI (aMCI) subjects and normal
control subjects to investigate the classification effectiveness of the memory contrasts
and subregions of medial temporal lobe. Least absolute shrinkage and selection
operator (LASSO) is more effective than principle component analysis (PCA) for
feature selection. The features selected by LASSO can be combined with
non-linear classifiers for high classification accuracy. The face-occupation
paradigm provides more discriminant power than the paradigms using outdoor
pictures or word pairs.
This abstract and the presentation materials are available to members only;
a login is required.