Hierarchical multi-scale dynamic graph analysis for early detection of change in EEG signals

The problem of automatic analysis and processing of electroencephalogram (EEG) signals has been an attractive but a quite difficult research subject with many potential clinical applications. For this problem, one major goal is to find and locate EEG state change that is from a normal state to an abnormal state during monitoring period, which remains challenging due to the high complexity of EEG signals. This study proposes a dynamic modeling method based on the combination of dynamic graph analysis and hierarchical decomposition (HD). The method can take into account the difference of contribution of individual frequency components in the EEG by construction of a family of hierarchical graph models (HGMs) from frequency components decomposed by HD. Adaptive input weighting (AIW) fusion is subsequently adopted to utilize and assemble these differences from all constructed graphs to form a comprehensive indicator that can describe the behavioral characteristics of EEG signals. Finally, a common null hypothesis testing is allowed for producing a decision making. Experiments results, our method can successfully detect all seizures without false alarm outperforming representable methods, demonstrated the availability and great potentials of the proposed method in real scenarios. This work is an extension of a recent result (Lu et al., 2020), which introduces an automated system that can detect changes in a given EEG time series. •Innovative Hierarchical Graph Modeling This approach combines dynamic graph analysis with hierarchical decomposition to address the complexity and diverse frequency components of EEG signals.•Adaptive Input Weighting Fusion Utilizing adaptive input weighting (AIW), this approach integrates the diverse contributions of individual frequency components into a comprehensive indicator.•Superior Performance and Validation Through experiments on two public datasets, this approach is proven to be a powerful clinical tool for early disease diagnosis and monitoring.