Machine learning identifies "rsfMRI epilepsy networks" in temporal lobe epilepsy.

OBJECTIVES: Experimental models have provided compelling evidence for the existence of neural networks in temporal lobe epilepsy (TLE). To identify and validate the possible existence of resting-state "epilepsy networks," we used machine learning methods on resting-state functional magnetic resonance imaging (rsfMRI) data from 42 individuals with TLE. METHODS: Probabilistic independent component analysis (PICA) was applied to rsfMRI data from 132 subjects (42 TLE patients + 90 healthy controls) and 88 independent components (ICs) were obtained following standard procedures. Elastic net-selected features were used as inputs to support vector machine (SVM). The strengths of the top 10 networks were correlated with clinical features to obtain "rsfMRI epilepsy networks." RESULTS: SVM could classify individuals with epilepsy with 97.5% accuracy (sensitivity = 100%, specificity = 94.4%). Ten networks with the highest ranking were found in the frontal, perisylvian, cingulo-insular, posterior-quadrant, thalamic, cerebello-thalamic, and temporo-thalamic regions. The posterior-quadrant, cerebello-thalamic, thalamic, medial-visual, and perisylvian networks revealed significant correlation (r > 0.40) with age at onset of seizures, the frequency of seizures, duration of illness, and a number of anti-epileptic drugs. CONCLUSIONS: IC-derived rsfMRI networks contain epilepsy-related networks and machine learning methods are useful in identifying these networks in vivo. Increased network strength with disease progression in these "rsfMRI epilepsy networks" could reflect epileptogenesis in TLE. KEY POINTS: • ICA of resting-state fMRI carries disease-specific information about epilepsy. • Machine learning can classify these components with 97.5% accuracy. • "Subject-specific epilepsy networks" could quantify "epileptogenesis" in vivo.

Machine learning detects EEG microstate alterations in patients living with temporal lobe epilepsy.

PURPOSE: Quasi-stable electrical distribution in EEG called microstates could carry useful information on the dynamics of large scale brain networks. Using machine learning techniques we explored if abnormalities in microstates can identify patients with Temporal Lobe Epilepsy (TLE) in the absence of an interictal discharge (IED). METHOD: 4 Classes of microstates were computed from 2 min artefact free EEG epochs in 42 subjects (21 TLE and 21 controls). The percentage of time coverage, frequency of occurrence and duration for each of these microstates were computed and redundancy reduced using feature selection methods. Subsequently, Fishers Linear Discriminant Analysis (FLDA) and logistic regression were used for classification. RESULT: FLDA distinguished TLE with 76.1% accuracy (85.0% sensitivity, 66.6% specificity) considering frequency of occurrence and percentage of time coverage of microstate C as features. CONCLUSION: Microstate alterations are present in patients with TLE. This feature might be useful in the diagnosis of epilepsy even in the absence of an IED.

NREM Sleep and Antiepileptic Medications Modulate Epileptiform Activity by Altering Cortical Synchrony.

INTRODUCTION: The activating role of non-rapid eye movement (NREM) sleep on epileptic cortex and conversely, the seizure remission brought about by antiepileptic medications, has been attributed to their effects on neuronal synchrony. This study aims to understand the role of neural synchrony of NREM sleep in promoting interictal epileptiform discharges (IEDs) in patients with epilepsy (PWE) by assessing the peri-IED phase synchrony during awake and sleep states. It also studies the role played by antiepileptic drugs (AEDs) on EEG desynchronization in the above cohort. METHODS: A total of 120 PWE divided into 3 groups (each n = 40; juvenile myoclonic epilepsy [JME], temporal lobe epilepsy [TLE]. and extratemporal lobe epilepsy [Ex-TLE]) were subjected to overnight polysomnography. Each patient group was subdivided into drug-naive and on treatment (Each n = 20). EEG phase synchronization analysis was performed to compare peri-IED phase synchronization indices (SI) during awake and sleep stages and between drug naïve and on treatment groups in 4 frequency bands, namely delta, theta, alpha, and beta. The mean ± SD of peri-IED SI among various subgroups was compared employing a multilevel mixed effects modeling approach. RESULTS: Patients with JME had increased peri-IED cortical synchrony in N3 sleep stage, whereas patients with partial epilepsy had increased IED cortical synchrony in N1 sleep stage. On the other hand, peri-IED synchrony was lower during wake and REM sleep. We also found that peri-IED synchronization in patients with JME was higher in drug-naive patients compared with those on sodium valproate monotherapy in theta, alpha, and beta bands. CONCLUSION: The findings of this study suggest that sleep stages can alter cortical synchrony in patients with JME and focal epilepsy, with NREM IEDs being more synchronized and wake/REM IEDs being less synchronized. Furthermore, it also suggests that AEDs alleviate seizures in PWE by inhibiting cortical synchrony.

Heightened Background Cortical Synchrony in Patients With Epilepsy: EEG Phase Synchrony Analysis During Awake and Sleep Stages Using Novel Ensemble Measure.

INTRODUCTION: Excessive cortical synchrony within neural ensembles has been implicated as an important mechanism driving epileptiform activity. The current study measures and compares background electroencephalographic (EEG) phase synchronization in patients having various types of epilepsies and healthy controls during awake and sleep stages. METHODS: A total of 120 patients with epilepsy (PWE) subdivided into 3 groups (juvenile myoclonic epilepsy [JME], temporal lobe epilepsy [TLE], and extra-temporal lobe epilepsy [Ex-TLE]; n = 40 in each group) and 40 healthy controls were subjected to overnight polysomnography. EEG phase synchronization (SI) between the 8 EEG channels was assessed for delta, theta, alpha, sigma, and high beta frequency bands using ensemble measure on 10-second representative time windows and compared between patients and controls and also between awake and sleep stages. Mean ± SD of SI was compared using 2-way analysis of variance followed by pairwise comparison ( P ≤ .05). RESULTS: In both delta and theta bands, the SI was significantly higher in patients with JME, TLE, and Ex-TLE compared with controls, whereas in alpha, sigma, and high beta bands, SI was comparable between the groups. On comparison of SI between sleep stages, delta band: progressive increase in SI from wake ⇒ N1 ⇒ N2 ⇒ N3, whereas REM (rapid eye movement) was comparable to wake; theta band: decreased SI during N2 and increase during N3; alpha band: SI was highest in wake and lower in N1, N2, N3, and REM; and sigma and high beta bands: progressive increase in SI from wake ⇒ N1 ⇒ N2 ⇒ N3; however, sigma band showed lower SI during REM. CONCLUSION: This study found an increased background cortical synchronization in PWE compared with healthy controls in delta and theta bands during wake and sleep. This background hypersynchrony may be an important property of epileptogenic brain circuitry in PWE, which enables them to effortlessly generate a paroxysmal EEG depolarization shift.

Phase Synchronization Analysis of Natural Wake and Sleep States in Healthy Individuals Using a Novel Ensemble Phase Synchronization Measure.

PURPOSE: Electrical activity in the brain is presumed to arise from a combination of tonic asynchronous neuronal firing during wake and a synchronized, burst-pause firing of large number of neurons during sleep. This study aims to compare the phase synchronization index (SI) across multiple channels during wake and various sleep stages on scalp electroencephalographic recordings. METHODS: Forty healthy subjects were subjected to overnight polysomnography using 8-channel electroencephalography. Electroencephalographic phase synchronization during awake, non-rapid eye movement (N1, N2, N3), and rapid eye movement sleep states was studied using ensemble measure (multichannel measure across all the eight channels based on Hilbert transformation between any two pairs). RESULTS: With the progression of states of wakefulness to non-rapid eye movement sleep, there was progressive increase in phase SI in delta band while SI decreased in alpha band (P < 0.001). The SI in delta band during rapid eye movement was comparable with that of awake state (P < 0.001). In theta band, SI tends to decrease in N2 and increase in N3 (P < 0.001). In beta band, there was progressive increase in SI from awake to non-rapid eye movement stages that decreased in rapid eye movement stage (P < 0.001). CONCLUSIONS: This is the first study that has used an ensemble measure to assess the long-range cortical phase synchronization during awake and various sleep stages. The findings support the previous view of increased delta synchrony during non-rapid eye movement sleep and alpha synchrony during wakefulness. Rapid eye movement stage was characterized by marked desynchrony in all frequency bands. These findings suggest the possible role of cortical synchronization in influencing the occurrence of epileptic activity during sleep and awake states.

Single-trial EEG classification using logistic regression based on ensemble synchronization.

In this paper, we propose an ensemble synchronization measure across all EEG channel pairs of a cluster based on Frobenius norm of the phase synchronization matrix, in a 0-1 scale enabling a direct comparison between clusters with different number of channels. Using this metric, we studied the intrahemispheric EEG synchronization in the lower gamma band (30-40 Hz) during 1229 single trials of an audio-visual integration cross modal task (CMT) recorded from five patients with schizophrenia and five healthy control subjects. Using ensemble synchronization measure and response latency of single trials recorded during the CMT as features for logistic regression, we could classify each single trial of EEG as belonging to a patient with schizophrenia or a healthy control subject with 73% accuracy, with an area under receiver operating characteristics curve of 0.83. We also propose a likelihood rating to denote the possibility of a subject belonging to the schizophrenia group.