Phase-amplitude coupling measures for determination of the epileptic network: A methodological comparison
Introduction
About 60% of people with epilepsy suffer from focal epilepsy (Rosenow and Lüders, 2001). In 30% of these patients, pharmacotherapy is insufficient and inefficient (Kwan et al., 2009, Ryvlin and Rheims, 2008). In these patients, epilepsy surgery can be a viable therapy option. Presurgical evaluation for epilepsy surgery helps to determine the epileptic focus or the epileptogenic zone (EZ) i.e., the minimum amount of brain tissue to be resected that will allow the patient to be seizure-free. The estimation of the EZ is purely theoretical and combines data obtained from multiple diagnostic tests such as scalp electro- and magneto-encephalography (EEG/MEG), positron emission tomography, single photon emission computed tomography, neuropsychological evaluation and high-resolution Magnetic Resonance Imaging (MRI) with epilepsy protocols. These techniques localize a potential epileptogenic lesion, but specifically also the electrophysiologically defined seizure onset and irritative zones, which serve as surrogate markers of the EZ.
The seizure onset zone (SOZ) which is defined as the region initiating ictal discharges during seizures, is considered to be one of the best markers of the EZ (Jehi, 2018). Invasive EEG or intracranial EEG (iEEG) i.e. electrocorticography (ECoG) or stereo-EEG (sEEG), is considered the gold standard for the localization of the SOZ and is considered mandatory in a significant portion of cases (Rosenow and Lüders, 2001, Ryvlin and Rheims, 2008).
However, with using ictal iEEG, the determination of SOZ relies on the recording of one, preferably several seizures. Therefore, this method is limited to patients with an adequate seizure frequency and even in this population, some patients might need to undergo iEEG recording for a couple of days to about a week to wait for a seizure to occur. A long recording period such as this may increase the risk of complications including infection and possible blood loss (Arya et al., 2013). As an alternative, interictal iEEG recordings are far easier to obtain. Interictal epileptic activity, such as spike and sharp wave patterns define the irritative zone (IZ), which may be identical or only overlap with the SOZ (Rosenow and Lüders, 2001). Previous studies have suggested that such data can act as a good measure to predict the SOZ as the cortical area producing interictal epileptic discharges (IEDs) overlaps with the seizure onset zone (Asano et al., 2003, Hufnagel et al., 2000).
Above and beyond IEDs, other candidate biomarkers that allow the portrayal of the SOZ or EZ in the interictal iEEG are high frequency oscillations (HFOs) (Bragin et al., 2010, Thomschewski et al., 2019). HFOs are divided into three subgroups: ripples (80–250 Hz), fast ripples (250–500 Hz), and very-fast ripples with frequencies exceeding 500 Hz (Bragin et al., 2010). It is suggested that HFOs occur at a higher rate within the SOZ or EZ than outside these areas (Andrade-Valenca et al., 2011, Jacobs et al., 2009, Jacobs et al., 2011, Sakuraba et al., 2016). Furthermore, the resection of areas with high rates of interictal HFOs is associated with a favorable surgical outcome (Iimura et al., 2018, Wang et al., 2017).
However, HFOs are also frequently generated by nonepileptic visual, somatosensory, motor, and auditory cortices during resting periods (Nagasawa et al., 2012, Wang et al., 2013). Therefore, it is probable that only some of the sites that generate interictal HFOs are associated with seizure generation (Asano et al., 2013). The literature currently does not recommend that the resection margin should solely be based on the occurrence rate or spectral frequency band of interictal HFOs (Asano et al., 2013; Engel et al., 2009).
Several recent studies explore the feasibility of using interactions between oscillations of different frequencies as plausible biomarkers for estimation of SOZ (Amiri et al., 2016, Motoi et al., 2018, Nonoda et al., 2016, Song et al., 2017). Oscillations of a particular frequency can modulate the oscillations in other frequency bands (Jensen and Colgin, 2007). The interaction between different frequency bands is called Cross-Frequency Coupling (CFC). Phase-amplitude coupling (PAC) describes the statistical dependence between the phase of a low-frequency oscillation and the amplitude (or power) of a high-frequency oscillation.
PAC has been observed in multiple regions of the human brain, including the visual cortex (Voytek et al., 2010), the auditory cortex (Cho et al., 2015), the hippocampus (Axmacher et al., 2010, Heusser et al., 2016) and prefrontal cortex (Voloh et al., 2015, Voytek et al., 2010) as well as the basal ganglia (Tort et al., 2008). Various functional roles of PAC have been suggested. For example, Voytek et al. (2010) reported that during visual tasks, the coupling between alpha and high-gamma oscillations increases preferentially in visual regions. They suggest that low-frequency to high-frequency coupling is modulated by behavioral tasks and that this may suggest a mechanism for selection between communicating neuronal networks. Similarly, Lakatos et al. (2008) discussed how delta-band oscillations in the visual cortex entrain to the rhythm of an applied stimulus and also determine the momentary power in the high-frequency activity which results in increased response gain to task-relevant events and decreased reaction times. Esghaei et al. (2015) found that spatial attention modulates the coupling between the low-frequency phase and high-frequency power. In particular, shifting attention to the receptive fields decreased PAC power for frequency pairs with the ‘phase-providing’ frequencies below 7 Hz. PAC between gamma and theta proved to be one of the most informative features in encoding information about the category of visual objects (Jafakesh et al., 2016).
Bonnefond et al. (2017) studied that PAC serves the dynamic coordination of brain activity over multiple spatial scales. Axmacher et al. (2010) suggested that maintenance of multiple items in working memory in the hippocampus is achieved by the coupling between gamma oscillations and consecutive phase ranges of oscillatory activity in the theta-frequency range. Cohen et al. (2009) studied how the positive and negative feedback processing was affected by the synchronization between the power in the alpha and beta bands and the phase of the delta and theta oscillations. Cohen et al. (2009) also showed that PAC between gamma and alpha increased substantially during reward-guided learning. It was also observed by Ahmadi et al. (2019) that PAC between the alpha-phase and the amplitude of frequencies 1–60 Hz generates half of the useful features in a Computer Aided Diagnostic system which diagnoses Multiple Sclerosis using information encoded during visual tasks.
PAC has also been investigated in patients with epilepsy. Multiple studies have shown that interictal HFOs in the seizure onset zone are coupled with the phase of slow waves (Amiri et al., 2019, Motoi et al., 2018, Nonoda et al., 2016, Ren et al., 2020). Ibrahim et al. (2014) have shown that PAC between HFO-amplitude and the phase of alpha and theta rhythms in the SOZ was significantly higher as compared to non-epileptic areas. Amiri et al. (2016) showed that PAC during different stages of sleep is significantly stronger in the SOZ than in the non-SOZ. Guirgis et al. (2015) observed that delta-modulated HFOs provided accurate localization of the EZ for the resection procedures in studied patients. In a previous study, we demonstrated that the degree of PAC is correlated to the density of highly active dysmorphic neurons in focal cortical dysplasia (FCD) II (Rampp et al., 2021).
Various methods have been used for the computation of PAC. The modulation index (MI) based on mean vector length (MVL) evaluates the dependence between the phase of a low-frequency oscillation and the amplitude of a high-frequency oscillation by the clustering of complex vectors (Canolty et al., 2006, Özkurt and Schnitzler, 2011). Another way to estimate coupling is to measure the non-uniformity of the distribution of the mean high-frequency amplitude across the bins of low-frequency phase. One implementation of this method has been done by Tort et al. (2010) and is known as the Kullback-Leibler distance-based modulation index (KL-MI). PAC can also be assessed by obtaining the mean over time of the phase-time series for the high-frequency oscillation and that for the low-frequency oscillation. This is known as the phase-locking value (PLV) (Penny et al., 2008) or synchronization index (SI) (Cohen, 2008).
It currently remains unclear as to which method performs best regarding the localization of SOZ although some comparisons have been made in the performance of some methods (Hülsemann et al., 2019, Jurkiewicz et al., 2020, Tort et al., 2010). Penny et al. (2008) compared PLV (Mormann et al., 2005), MVL-MI (Canolty et al., 2006) and envelope-to-signal correlation (ESC) (Bruns and Eckhorn, 2004) to the generalized linear modeling (GLM) method. They considered the impact of factors such as data length, amount of noise, sampling rate etc. They concluded that all methods performed on the same level when the conditions were suitable e.g., longer epochs and low noise. When these factors were varied, the performance differed.
Here we compare four methods namely MVL-MI (Canolty et al., 2006), the direct MVL-MI estimator (Özkurt and Schnitzler, 2011), PLV (Cohen, 2008) and the KL-MI (Tort et al., 2008, Tort et al., 2010) in terms of SOZ estimation. These methods are used to compute PAC between the amplitude of the gamma frequency and ripples and the phases of alpha, theta and delta oscillations. We also investigate whether the choice of different phase-frequencies has any impact on the PAC.
Section snippets
Participants
We investigated a series of fifteen consecutive patients who all suffered from pharmaco-resistant focal epilepsy due to suspected focal cortical dysplasia. FCD is the most common cause of medically refractory epilepsy in the pediatric population and the second/third most common etiology of medically intractable seizures in adults (Kabat and Król, 2012). This selection criterion was chosen as FCDs are considered to be highly epileptogenic and serve here as the definition of the epileptic focus.
Gamma-PAC
The Shapiro-Wilk test on the interictal, ictal as well the related AUC values showed a non-normal distribution, also after log-transformation (Asano et al., 2009). A one-way non-parametric Kruskal-Wallis H-test showed that the factor ‘Method’ (i.e., MVL-MI-Canolty, dMVL-MI-Özkurt, KL-MI-Tort, PLV-Cohen) significantly affects the PAC for interictal H(3) = 18.83, p = 0.0003, as well as for ictal H(3) = 10.58, p = 0.0142 and combined H(3) = 20.97, p = 0.0001. The Kruskal Wallis H-test did not
PAC as a marker of SOZ
In our data, PAC demonstrated the ability to identify the EZ, in line with the literature. For example, Weiss et al. (2015) showed that HFOs phase-locked to lower frequency IEDs are associated with the SOZ. They argue that the coupled IEDs might reflect altered excitability giving rise to abnormally synchronized and hyperexcitable cell assemblies as a potential source for HFO activity. In addition to previous data on HFOs in epilepsy (Frauscher et al., 2017), their observation implies
Conclusion
In this study we showed that PAC of continuous interictal data can be used to detect both SOZ and IZ. Algorithm choice influenced the ability to identify the SOZ and IZ when analyzing gamma- but not ripple-PAC. AUC values for gamma-PAC were highest with MVL-MI-Canolty, followed by KL-MI-Tort. dMVL-MI-Özkurt and PLV-Cohen showed considerably lower performance.
The choice of the lower phase-frequency band did not significantly affect computation results for gamma-PAC. Ripple-PAC, however,
CRediT authorship contribution statement
Ryshum Ali: Formal Analysis, Software, Writing – original draft preparation, Visualization, Stephanie Gollwitzer: Investigation, Caroline Reindl: Investigation, Hajo Hamer: Resources, Investigation, Roland Coras: Investigation, Ingmar Blümcke: Funding Acquisition, Resources, Investigation, Michael Buchfelder: Resources, Peter Hastreiter: Supervision, Stefan Rampp: Conceptualization, Supervision, Writing – review & editing.
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Acknowledgments
This work was supported by the European Union's Seventh Framework Programme for research, technological development and demonstration (EU FP7) grant EU-HEALTH-2013, DESIRE, grant agreement no. 602531 to Ingmar Blümcke.
References (69)
- et al.
Computer aided diagnosis system for multiple sclerosis disease based on phase to amplitude coupling in covert visual attention
Comput. Methods Prog. Biomed.
(2019) - et al.
Task-related coupling from high- to low-frequency signals among visual cortical areas in human subdural recordings
Int. J. Psychophysiol.
(2004) Assessing transient cross-frequency coupling in EEG data
J. Neurosci. Methods
(2008)- et al.
Dynamic modulation of epileptic high frequency oscillations by the phase of slower cortical rhythms
Exp. Neurol.
(2014) - et al.
Decoding of object categories from brain signals using cross frequency coupling methods
Biomed. Signal Process. Control
(2016) - et al.
Cross-frequency coupling between neuronal oscillations
Trends Cogn. Sci.
(2007) - et al.
A critical note on the definition of phase–amplitude cross-frequency coupling
J. Neurosci. Methods
(2011) - et al.
Testing for nested oscillation
J. Neurosci. Methods
(2008) - et al.
Defining the filter parameters for phase-amplitude coupling from a bispectral point of view
J. Neurosci. Methods
(2021) - et al.
Phase-amplitude coupling is elevated in deep sleep and in the onset zone of focal epileptic seizures
Front. Hum. Neurosci.
(2016)