Learning Synchronization Patterns in Neural Signal

Wider research context/theoretical framework
We will use the concept of synchronization of nonlinear dynamical systems as the basis for development of data mining methods and algorithms for detection and modelling of interactions and dependence patterns in multivariate nonlinear time series.

Hypotheses/research questions/objectives
The developed methods will be tailored to specific properties of scalp electroencephalogram (EEG), in order to better understand synchronization phenomena in the human brain. The overall structure of EEG synchronization, reflecting the functional integration of the brain activity within and across different spatial and temporal scales, will be characterized within the framework of complex networks and graph theory. These formal tools will be used for description of brain states and their changes in mental disorders. In particular, synchronization and its changes in EEG of depressive patients will be tested as predictors of antidepressant therapeutic efficacy. Due to the formal approaches proposed in the project, the developed methods will be applicable not only in analysis of electrophysiological signals in psychiatry, but generally in analysis of complex multivariate and multiscale signals.

Approach/Methods
The machine learning and data mining tools will have a twofold role in the project. First, the synchronization patterns in the form of complex, multilayer networks will be used as an input of a learning algorithm in order to find features characterizing different physiological and/or pathological brain states. In parallel, the synchronization principle will be used as a basis for data mining. The synchronization-based clustering (Austrian PI) will be applied directly to EEG time series in order to obtain groups of subjects characterized by their interaction patterns reflected by suitable models of EEG (phase) synchronization. Both approaches will be used in the proposed primary application for extracting predictors of antidepressant therapeutic efficacy from multichannel EEG.

Level of originality/innovation
The application promises high impact for society, since depression is a major cause of morbidity worldwide. According to WHO, the number of people suffering from depression increased by 18% in 2005- 2015. In EU countries the occurrence of depressive disorders counts between 2.6 – 4.5% for males and 7.1 – 10.4% for females. Modern antidepressant drugs have a response rate only up to 65%, whereas the response requires usually 4–6 weeks of treatment. The ability to predict response to treatment, either early in the course of therapy or before treatment even begins, can save patients from prolonged intervals of suffering. Also economic aspects of early choice of an effective therapy are indisputable.

We analyzed electrical brain activity recordings (EEG) from patients on day 7 of antidepressant treatment. We tested different ways of describing these EEG signals, i.e. we extracted different features from the EEG recordings primarily capturing synchronization patterns between brain regions, and trained machine learning models to learn the difference between people who later improved and those who did not. In addition, we explored whether patients can be grouped into meaningful subtypes based on similarities in their brain activity patterns—an approach that may help explain why the same treatment works well for some people but not for others.

The strongest results came from focusing on short, recurring signal patterns in the EEG (a “motif”-based approach). Using this method, we correctly predicted treatment outcome for 73% of patients in an independent validation group. This is a key scientific advance: it demonstrates that early EEG patterns can contain practical information about later treatment success, and it identifies a particularly effective way to extract that information.

If confirmed in larger studies, this approach could support earlier, more personalized treatment decisions—helping patients reach an effective therapy sooner, reducing avoidable side effects and suffering, and potentially saving healthcare resources.