Calculating dynamic connectivity around EEG seizure onset based on nonlinearity and nonstationarity
Poster No:
1528
Submission Type:
Abstract Submission
Authors:
Iris Soare-Nguyen1,2, Shima Abdullateef3, Afreen Islam4, Javier Escudero1,2
Institutions:
1Institute for Imaging, Data and Communications, School of Engineering, University of Edinburgh, Edinburgh, Midlothian, 2Muir Maxwell Epilepsy Centre, Edinburgh, Midlothian, United Kingdom, 3Usher Institute, University of Edinburgh, Edinburgh, Midlothian, 4Control and Robotics Group, Dep. of Electrical and Electronic Engineering, University of Manchester, Manchester, Greater Manchester
First Author:
Iris Soare-Nguyen
Institute for Imaging, Data and Communications, School of Engineering, University of Edinburgh|Muir Maxwell Epilepsy Centre
Edinburgh, Midlothian|Edinburgh, Midlothian, United Kingdom
Institute for Imaging, Data and Communications, School of Engineering, University of Edinburgh|Muir Maxwell Epilepsy Centre
Edinburgh, Midlothian|Edinburgh, Midlothian, United Kingdom
Co-Author(s):
Afreen Islam
Control and Robotics Group, Dep. of Electrical and Electronic Engineering, University of Manchester
Manchester, Greater Manchester
Control and Robotics Group, Dep. of Electrical and Electronic Engineering, University of Manchester
Manchester, Greater Manchester
Javier Escudero
Institute for Imaging, Data and Communications, School of Engineering, University of Edinburgh|Muir Maxwell Epilepsy Centre
Edinburgh, Midlothian|Edinburgh, Midlothian, United Kingdom
Institute for Imaging, Data and Communications, School of Engineering, University of Edinburgh|Muir Maxwell Epilepsy Centre
Edinburgh, Midlothian|Edinburgh, Midlothian, United Kingdom
Introduction:
Seizures represent abnormal brain activity due to unregulated synchronization. Connectivity can measure spatiotemporal interaction between brain network nodes to characterize such synchronization.
Brain activity transitions between emergent patterns that can be locally characterized as linear or stationary, interrupted by nonlinear bursts (Roberts et al. 2019; Breakspear et al. 2006). Supplementarily, seizures are known to be strongly nonlinear (Lancaster et al. 2018) and nonstationary (Manuca et al. 1998). Thus, testing linearity and stationarity of epileptic electroencephalograms (EEGs) in the context of connectivity is of interest (Zhang et al. 2020), as shifts in signal properties introduce errors in connectivity estimation. Changes in nonlinearity signify shifts in emergence (Roberts et al. 2019), and segments may also exhibit local stationarity changes around seizures (Borgnat et al. 2010).
To explore epileptogenic network evolution, we compute delta band dynamic effective connectivity (dEC) on human epileptic EEGs. By adopting a piecewise first order approximation across multiple points in the system's trajectory (Leith and Leithead 1999), we inspect the changes in linearity and stationarity to differentiate states. We summarize the signal and connectivity properties we found.
Brain activity transitions between emergent patterns that can be locally characterized as linear or stationary, interrupted by nonlinear bursts (Roberts et al. 2019; Breakspear et al. 2006). Supplementarily, seizures are known to be strongly nonlinear (Lancaster et al. 2018) and nonstationary (Manuca et al. 1998). Thus, testing linearity and stationarity of epileptic electroencephalograms (EEGs) in the context of connectivity is of interest (Zhang et al. 2020), as shifts in signal properties introduce errors in connectivity estimation. Changes in nonlinearity signify shifts in emergence (Roberts et al. 2019), and segments may also exhibit local stationarity changes around seizures (Borgnat et al. 2010).
To explore epileptogenic network evolution, we compute delta band dynamic effective connectivity (dEC) on human epileptic EEGs. By adopting a piecewise first order approximation across multiple points in the system's trajectory (Leith and Leithead 1999), we inspect the changes in linearity and stationarity to differentiate states. We summarize the signal and connectivity properties we found.
Methods:
We studied 20 human epileptic EEG recordings (Detti 2020) consisting of five minutes of preictal and then ictal activity. We preprocessed the data with a low-pass Butterworth filter with cut-off at 45 Hz and removing blinks and saccades.
We tested stationarity separately on every channel by computing the Euclidean distance of spectrograms between every 1s segment pair. We computed the cumulative median absolute deviation (MAD) z-score across each channel (Edwards et al. 2020). Segments with a lower MAD than 1 were considered stationary.
To test linearity for each 1s segment, we created 40 iterative amplitude adjusted Fourier transform (IAAFT2) surrogates (Lancaster et al. 2018), then computed the nonlinear measure of Lyapunov exponent (LE) of the signal and surrogates. We considered all segments with a lower LE than 97.5% of the surrogates as linear.
Finally, we fed the stationarity and linearity scores into k-medoids clustering to obtain consecutive signal segments with comparable properties. For each cluster of segments, we computed dEC using spectrally resolved Granger causality. We obtained proof-of-concept delta band dynamic connectivity for each seizure.
We tested stationarity separately on every channel by computing the Euclidean distance of spectrograms between every 1s segment pair. We computed the cumulative median absolute deviation (MAD) z-score across each channel (Edwards et al. 2020). Segments with a lower MAD than 1 were considered stationary.
To test linearity for each 1s segment, we created 40 iterative amplitude adjusted Fourier transform (IAAFT2) surrogates (Lancaster et al. 2018), then computed the nonlinear measure of Lyapunov exponent (LE) of the signal and surrogates. We considered all segments with a lower LE than 97.5% of the surrogates as linear.
Finally, we fed the stationarity and linearity scores into k-medoids clustering to obtain consecutive signal segments with comparable properties. For each cluster of segments, we computed dEC using spectrally resolved Granger causality. We obtained proof-of-concept delta band dynamic connectivity for each seizure.
Results:
As Fig. 1A and 1C show, we consistently found a change in stationarity with seizure onset. However, the median value of all previous Euclidean distances introduced a delay in identifying the change in channel signal stationarity.
The linearity results indicate complex dynamics: nonlinear segments are spread across every seizure; at any moment, certain channels show nonlinear LE scores while others do not. This confirms that brain activity is mostly linear with intermittent nonlinear fluctuations (Breakspear et al. 2006). By taking the linear fit of the mean LE scores we also show that ictal activity displays either decreasing or increasing nonlinearity compared to preictal activity, as Fig. 1D shows. This variability between seizures represents future work.
Our k-medoids results indicate a trend of faster network reorganization before than during seizure with a p-value of 0.061. This is in agreement with (Khambhati et al. 2015), as Fig. 2 shows. Unlike them, we were not able to distinguish three ictal network reconfigurations, which requires further investigation.
The linearity results indicate complex dynamics: nonlinear segments are spread across every seizure; at any moment, certain channels show nonlinear LE scores while others do not. This confirms that brain activity is mostly linear with intermittent nonlinear fluctuations (Breakspear et al. 2006). By taking the linear fit of the mean LE scores we also show that ictal activity displays either decreasing or increasing nonlinearity compared to preictal activity, as Fig. 1D shows. This variability between seizures represents future work.
Our k-medoids results indicate a trend of faster network reorganization before than during seizure with a p-value of 0.061. This is in agreement with (Khambhati et al. 2015), as Fig. 2 shows. Unlike them, we were not able to distinguish three ictal network reconfigurations, which requires further investigation.
Conclusions:
We computed dEC of human epileptic EEG activity by clustering signal segments in terms of changes in stationarity and linearity. The signal segmentation suggests that the epileptogenic network changes more rapidly and displays different behavior in terms of linearity and stationarity around seizures. We will use the obtained dynamic connectivity to study linear time-variant network control strategies.
Disorders of the Nervous System:
Neurodevelopmental/ Early Life (eg. ADHD, autism) 2
Modeling and Analysis Methods:
Connectivity (eg. functional, effective, structural) 1
EEG/MEG Modeling and Analysis
Keywords:
Electroencephaolography (EEG)
Epilepsy
Other - Dynamic connectivity, Nonlinearity, Nonstationarity
1|2Indicates the priority used for review
Provide references using author date format
Breakspear, M., J. A. Roberts, J. R. Terry, S. Rodrigues, N. Mahant, and P. A. Robinson. 2006. “A Unifying Explanation of Primary Generalized Seizures Through Nonlinear Brain Modeling and Bifurcation Analysis.” Cerebral Cortex 16 (9): 1296–1313. https://doi.org/10.1093/cercor/bhj072.
Detti, P.. 2020. “Siena Scalp EEG Database.” PhysioNet. https://doi.org/10.13026/5D4A-J060.
Edwards, M. C., P. Maturana-Russel, R. Meyer, J. Gair, N. Korsakova, and N. Christensen. 2020. “Identifying and Addressing Nonstationary LISA Noise.” Physical Review D 102 (8): 084062. https://doi.org/10.1103/PhysRevD.102.084062.
Khambhati, A. N., K. A. Davis, B. S. Oommen, S. H. Chen, T. H. Lucas, B. Litt, and D. S. Bassett. 2015. “Dynamic Network Drivers of Seizure Generation, Propagation and Termination in Human Neocortical Epilepsy.” PLOS Computational Biology 11 (12): e1004608. https://doi.org/10.1371/journal.pcbi.1004608.
Lancaster, G., D. Iatsenko, A. Pidde, V.a Ticcinelli, and A. Stefanovska. 2018. “Surrogate Data for Hypothesis Testing of Physical Systems.” Physics Reports, Surrogate data for hypothesis testing of physical systems, 748 (July): 1–60. https://doi.org/10.1016/j.physrep.2018.06.001.
Leith, D J, and W. E. Leithead. 1999. “Input-Output Linearisation by Velocity-Based Gain-Scheduling.” International Journal of Control 72 (3). https://www.hamilton.ie/doug/1998-3.pdf.
Manuca, R., M. C. Casdagli, and R. S. Savit. 1998. “Nonstationarity in Epileptic EEG and Implications for Neural Dynamics.” Mathematical Biosciences 147 (1): 1–22. https://doi.org/10.1016/S0025-5564(97)00055-2.
Roberts, J. A., L. L. Gollo, R. G. Abeysuriya, G. Roberts, P. B. Mitchell, M. W. Woolrich, and M. Breakspear. 2019. “Metastable Brain Waves.” Nature Communications 10 (1): 1056. https://doi.org/10.1038/s41467-019-08999-0.
Zhang, Y., D. Lai, J. Han, X. Wang, Q. Lin, X. Zhao, and Z. Hu. 2020. “Testing Nonlinearity in Topological Organization of Functional Brain Networks.” ” European Journal of Neuroscience 52 (9): 4185–97. https://doi.org/10.1111/ejn.14882.