Dynamic brain effective connectivity network for identifying neurological disorders

Saqib Mamoon, Zhengwang Xia, Amani Alfakih, Jianfeng Lu

Abstract: Brain dynamic Effective Connectivity (dEC) captures the time-varying causal interactions among the brain regions, offering valuable insights into neural mechanisms. However, prevailing methods primarily focus on static or temporally invariant Effective Connectivity (EC), leaving a significant knowledge gap regarding dEC’s dynamic aspect. Moreover, the current methods either require strong prior assumptions about underlying brain connectivity or rely on predefined thresholds when modeling the brain network. Herein, we propose a novel neural network-based Dynamic Brain Effective Connectivity Network (DB-ECN) approach that infers the EC at each time step, characterizing the temporal evolution of connectivity patterns in the brain. Notably, our method incorporates two key features: a second-order smoothing mechanism to suppress abrupt changes and a reasonable aggregation scheme to obtain an overall brain representation. Moreover, a new thresholding scheme is introduced to adaptively select the most reliable brain connections. Our method is highly interpretable, highlighting the transition from normal to abnormal brain connectivities and facilitating the identification of potential biomarkers. The experimental results on two large-scale public datasets demonstrate that DB-ECN is superior to other state-of-the-art methods in learning the network structure and distinguishing abnormal from normal patterns of brain activities. Our proposed methodology offers promising insights into brain dynamics and disorder diagnosis, paving the way for advancements in neuroscience research and clinical applications.

External IDs:dblp:journals/apin/MamoonXAL25