COOPERATIVE DIFFERENTIAL NETWORK LEARNING WITH HUB DETECTION FOR MULTICENTER NEUROIMAGING DATA

In this study we focus on Cooperative Differential Network Learning with hub detection (CDNL) for functional Magnetic Resonance Imaging (fMRI) scan from multiple research centers. As research centers may use varying scanners, imaging parameters, and other conditions that introduce heterogeneity, CDNL allows us to analyze fMRI data from various perspectives while identifying shared structures, potentially revealing the underlying mechanisms of neurological diseases. In addition, brain functional networks often consist of multiple hubs-central nodes within the network that play a crucial role in supporting integrated brain function. Investigating these hubs can offer valuable insight into functional connectivity patterns in the brain. To address this task, we formulate it as a penalized logistic regression problem and introduce two independent penalties (Cooperative Penalty and Hub Penalty) to enable simultaneous estimation of multiple differential networks with hub detection. To further enhance empirical performance, we develop an ensemble-learning procedure. We conduct comprehensive simulation studies to assess the finite-sample performance of the proposed method and compare it with existing state-of-the-art alternatives. In the application we apply the proposed method to analyze multiple fMRI scans related to Attention Deficit Hyperactivity Disorder from various research centers. We identify common hub brain regions and similar differential interaction patterns across various centers. These findings are highly consistent with existing results from clinical medical research.