A0191
Title: Improving the reproducibility of brain imaging feature selection with weighted regularization
Authors: Fengqing Zhang - Drexel University (United States) [presenting]
Abstract: Advances in brain imaging techniques and machine learning (ML) models allow researchers to combine many brain imaging features to aid the diagnosis of psychiatric disorders jointly. Increasingly, the reproducibility of ML results has drawn great attention. Studies examining the reproducibility problem in brain imaging have largely focused on prediction accuracy. However, achieving high prediction accuracy and discovering relevant features are not necessarily the same. An important yet under-investigated problem is the reproducibility of feature selection in brain imaging studies. A new metric is proposed to quantify the reproducibility of neuroimaging feature selection via bootstrapping. The reproducibility index (R-index) is estimated for each feature as the reciprocal coefficient of variation of absolute mean difference across a larger number of bootstrap samples. The R-index in regularized classification models is then integrated as penalty weight. Reproducible features with a larger R-index are assigned smaller penalty weights and, thus, are more likely to be selected by the proposed models. The performance of the proposed models is evaluated using both simulated and multimodal neuroimaging data.
