Xiaowei Zhuang1, Virendra Mishra1, Karthik Sreenivasan1, Charles Bernick1, Sarah Banks1, and Dietmar Cordes1,2
1Cleveland Clinic Lou Ruvo Center for Brain Health, Las Vegas, NV, United States, 2Department of Psychology and Neuroscience, University of Colorado Boulder, Boulder, CO, United States
Clinical
applications of brain abnormality detection with supervised machine learning
techniques are limited due to less and
unbalanced sample sizes as compared to rich feature sets
in patient population. We proposed a new combinatorial model approach, fs-RBFN,
involving sampling from multivariate joint distribution, LASSO feature
selection, RBFN cross validation, and inverse probability weighting to solve
this problem. The proposed approach was validated against a ground truth
phantom and further tested on a multimodal MRI dataset for cognitively impaired
and non-impaired professional fighters. Our results suggest superior
performance of this technique over several other out-of-the-bag feature
selection algorithms.
