Defining regions of interest using cross-frequency coupling in extratemporal lobe epilepsy patients.

OBJECTIVE: Clinicians identify seizure onset zones (SOZs) for resection in an attempt to localize the epileptogenic zone (EZ), which is the cortical tissue that is indispensible for seizure generation. An automated system is proposed to objectively localize this EZ by identifying regions of interest (ROIs). METHODS: Intracranial electroencephalogram recordings were obtained from seven patients presenting with extratemporal lobe epilepsy and the interaction between neuronal rhythms in the form of phase-amplitude coupling was investigated. Modulation of the amplitude of high frequency oscillations (HFOs) by the phase of low frequency oscillations was measured by computing the modulation index (MI). Delta- (0.5-4 Hz) and theta- (4-8 Hz) modulation of HFOs (30-450 Hz) were examined across the channels of a 64-electrode subdural grid. Surrogate analysis was performed and false discovery rates were computed to determine the significance of the modulation observed. Mean MI values were subjected to eigenvalue decomposition (EVD) and channels defining the ROIs were selected based on the components of the eigenvector corresponding to the largest eigenvalue. ROIs were compared to the SOZs identified by two independent neurologists. Global coherence values were also computed. MAIN RESULTS: MI was found to capture the seizure in time for six of seven patients and identified ROIs in all seven. Patients were found to have a poorer post-surgical outcome when the number of EVD-selected channels that were not resected increased. Moreover, in patients who experienced a seizure-free outcome (i.e., Engel Class I) all EVD-selected channels were found to be within the resected tissue or immediately adjacent to it. In these Engel Class I patients, delta-modulated HFOs were found to identify more of the channels in the resected tissue compared to theta-modulated HFOs. However, for the Engel Class IV patient, the delta-modulated HFOs did not identify any of the channels in the resected tissue suggesting that the resected tissue was not appropriate, which was also suggested by the Engel Class IV outcome. A sensitivity of 75.4% and a false positive rate of 15.6% were achieved using delta-modulated HFOs in an Engel Class I patient. SIGNIFICANCE: LFO-modulated HFOs can be used to identify ROIs in extratemporal lobe patients. Moreover, delta-modulated HFOs may provide more accurate localization of the EZ. These ROIs may result in better surgical outcomes when used to compliment the SOZs identified by clinicians for resection.

Classification of Multiple Seizure-Like States in Three Different Rodent Models of Epileptogenesis.

Epilepsy is a dynamical disease and its effects are evident in over fifty million people worldwide. This study focused on objective classification of the multiple states involved in the brain's epileptiform activity. Four datasets from three different rodent hippocampal preparations were explored, wherein seizure-like-events (SLE) were induced by the perfusion of a low - Mg(2+) /high-K(+) solution or 4-Aminopyridine. Local field potentials were recorded from CA3 pyramidal neurons and interneurons and modeled as Markov processes. Specifically, hidden Markov models (HMM) were used to determine the nature of the states present. Properties of the Hilbert transform were used to construct the feature spaces for HMM training. By sequentially applying the HMM training algorithm, multiple states were identified both in episodes of SLE and nonSLE activity. Specifically, preSLE and postSLE states were differentiated and multiple inner SLE states were identified. This was accomplished using features extracted from the lower frequencies (1-4 Hz, 4-8 Hz) alongside those of both the low- (40-100 Hz) and high-gamma (100-200 Hz) of the recorded electrical activity. The learning paradigm of this HMM-based system eliminates the inherent bias associated with other learning algorithms that depend on predetermined state segmentation and renders it an appropriate candidate for SLE classification.

Gamma (30-80Hz) bicoherence distinguishes seizures in the human epileptic brain.

We have applied wavelet bicoherence (BIC) analysis to human iEEG data to characterize non-linear frequency interactions in the human epileptic brain. Bicoherence changes were most prominent in the gamma (30-80 Hz) frequency band, and allowed for the differentiation between seizure and non-seizure states in all patients studied (n=3). While gamma band BIC values increased during seizure activity, this trend was only observed in a select number of electrode(s) located on the implanted patient subdural grids. Several studies have suggested that fast frequencies may play a role in the process of seizure genesis. While the small patient numbers limit the significance of our study, our results highlight the bicoherence of the gamma frequency band (30-80 Hz) as an ictal identifier, and suggest an active role of this fast frequency during seizures.

The role of delta-modulated high frequency oscillations in seizure state classification.

High frequency oscillations (HFOs), which collectively refer to ripples (80-200 Hz) and fast ripples (>200 Hz), have been implicated as key players in epileptogenesis. However, their presence alone is not in and of itself indicative of a pathological brain state. Rather, spatial origins as well as coexistence with other neural rhythms are essential components in defining pathological HFOs. This study investigates how the phase of the delta rhythm (0.5-4 Hz) modulates the amplitude of HFOs during a seizure episode. Seven seizures recorded from three patients presenting with intractable temporal lobe epilepsy were obtained via intracranial electroencephalography (iEEG) from a 64-electrode grid. Delta modulation of the HFO rhythms was found to emerge at seizure onset and termination regardless of the dynamics present within the seizure episode itself. Moreover, the differences between delta modulating the ripple or fast ripple may be due to the sleep stage of the patient when the seizures were being recorded. Further studies exploring how this modulation changes in space across the grid may also highlight additional properties of this phenomenon. Its temporal pattern suggests that it is a potential iEEG-based biomarker for seizure state classification.

Taking NIRS-BCIs outside the lab: towards achieving robustness against environment noise.

This paper reported initial findings on the effects of environmental noise and auditory distractions on the performance of mental state classification based on near-infrared spectroscopy (NIRS) signals recorded from the prefrontal cortex. Characterization of the performance losses due to environmental factors could provide useful information for the future development of NIRS-based brain-computer interfaces that can be taken beyond controlled laboratory settings and into everyday environments. Experiments with a hidden Markov model-based classifier showed that while significant performance could be attained in silent conditions, only chance levels of sensitivity and specificity were obtained in noisy environments. In order to achieve robustness against environment noise, two strategies were proposed and evaluated. First, physiological responses harnessed from the autonomic nervous system were used as complementary information to NIRS signals. More specifically, four physiological signals (electrodermal activity, skin temperature, blood volume pulse, and respiration effort) were collected in synchrony with the NIRS signals as the user sat at rest and/or performed music imagery tasks. Second, an acoustic monitoring technique was proposed and used to detect startle noise events, as both the prefrontal cortex and ANS are known to involuntarily respond to auditory startle stimuli. Experiments with eight participants showed that with a startle noise compensation strategy in place, performance comparable to that observed in silent conditions could be recovered with the hybrid ANS-NIRS system.

Capturing the state transitions of seizure-like events using Hidden Markov models.

The purpose of this study was to investigate the number of states present in the progression of a seizure-like event (SLE). Of particular interest is to determine if there are more than two clearly defined states, as this would suggest that there is a distinct state preceding an SLE. Whole-intact hippocampus from C57/BL mice was used to model epileptiform activity induced by the perfusion of a low Mg(2+)/high K(+) solution while extracellular field potentials were recorded from CA3 pyramidal neurons. Hidden Markov models (HMM) were used to model the state transitions of the recorded SLEs by incorporating various features of the Hilbert transform into the training algorithm; specifically, 2- and 3-state HMMs were explored. Although the 2-state model was able to distinguish between SLE and nonSLE behavior, it provided no improvements compared to visual inspection alone. However, the 3-state model was able to capture two distinct nonSLE states that visual inspection failed to discriminate. Moreover, by developing an HMM based system a priori knowledge of the state transitions was not required making this an ideal platform for seizure prediction algorithms.

On the use of peripheral autonomic signals for binary control of body-machine interfaces.

In this work, the potential of using peripheral autonomic (PA) responses as control signals for body-machine interfaces that require no physical movement was investigated. Electrodermal activity, skin temperature, heart rate and respiration rate were collected from six participants and hidden Markov models (HMMs) were used to automatically detect when a subject was performing music imagery as opposed to being at rest. Experiments were performed under controlled silent conditions as well as in the presence of continuous and startle (e.g. door slamming) ambient noise. By developing subject-specific HMMs, music imagery was detected under silent conditions with the average sensitivity and specificity of 94.2% and 93.3%, respectively. In the presence of startle noise stimuli, the system sensitivity and specificity levels of 78.8% and 80.2% were attained, respectively. In environments corrupted by continuous ambient and startle noise, the system specificity further decreased to 75.9%. To improve the system robustness against environmental noise, a startle noise detection and compensation strategy were proposed. Once in place, performance levels were shown to be comparable to those observed in silence. The obtained results suggest that PA signals, combined with HMMs, can be useful tools for the development of body-machine interfaces that allow individuals with severe motor impairments to communicate and/or to interact with their environment.