Phenotype Prediction Using BEN-Based Predictive Modeling (BPM)

Abstract

Functional connectivity (FC) has been successfully used to predict cognitive functions, behaviors, and other phenotypes using connectome-based predictive modeling (CPM). One limitation is that FC reflects the covariation or synchrony relationship of the temporal profile in two regions and neglects the local temporal features of the brain activity. In this study, we used brain entropy (BEN) mapping to characterize regional brain activity and developed a BEN-Based Predictive Modeling (BPM) to predict phenotype. BEN measures the disorder and complexity of brain activity and has been shown to effectively capture brain activity features related to cognition and neurological disorders. Using data from the HCP 7T fMRI and corresponding 3T structural images, we calculated gray matter volume (GMV) as well as BEN from resting state and movie-watching data. We constructed prediction models based on BEN and GMV using different numbers of parcellation of the brain atlas, applying 10-fold cross-validation. Our results indicated that the BEN-based predictive model not only outperformed GMV-based predictive modeling but also achieved prediction accuracy comparable to that of CPM. Our study demonstrates that BEN can capture extensive brain activity information for accurate phenotype prediction, providing information at least as valuable as that from connectome. Additionally, our research lays the groundwork for future applications of BPM in developmental and clinical practice.