Research

OBJECTIVE: Approximately 40% of children undergoing epilepsy surgery have postoperative seizures, underscoring the need for enhanced estimators of the epileptogenic zone (EZ). We hypothesize that visually imperceptible low-entropy activity in the interictal periods, even in the absence of conventional spikes, is a robust signature of the EZ. To test this, we mapped interictal "low-entropy zones" using intracranial electroencephalography (iEEG) in children with drug-resistant epilepsy (DRE) and assessed their value for postsurgical outcome prediction when targeted during surgery, along with their stability over prolonged periods.

METHODS: We analyzed iEEG data of 75 DRE children, including brief (5 min) data from patients with known Engel outcome (N = 59; used for outcome prediction) plus prolonged data from a separate recent cohort (N = 16; used for stability assessment). We estimated each contact's entropy across various frequencies (delta to fast-ripple), pinpointed low-entropy zones, and assessed whether their removal predicts outcome (3-fold cross-validation). In addition, the predictive value of entropy during non-epileptiform (spike-free) epochs was also assessed. Furthermore, established interictal estimators (spikes-on-ripple, fast ripples) were tested for outcome prediction. Using the prolonged dataset, we tested whether entropy distribution over brief epochs was similar to prolonged (3 h) data.

RESULTS: High overlap between low-entropy zones and resection correlated with low Engel class (p < 0.0001, R = -0.54, N = 59), also during non-epileptiform epochs (R = -0.52). Low-entropy-zone removal predicted outcomes with F1 score of 87% (p < 0.0001, N = 51; Engel I vs III-IV) outperforming spikes-on-ripple (F1 score = 82%, p = 0.002) or fast ripples (F1 score = 80%, p = 0.01). Low-entropy zones retained high predictive value when non-epileptiform epochs were used (F1 score = 89%, N = 44). Entropy distribution over brief epochs was strongly correlated with prolonged data (R > 0.8, p < 0.0001), and its relationship with seizure-onset zone did not differ (brief vs prolonged data: p > 0.6).

SIGNIFICANCE: Surgically targeting low-entropy zones accurately predicts the postoperative seizure outcomes of children with DRE. Mapping low-entropy activity using brief iEEG segments shows consistency with using prolonged data and could enhance surgical planning in pediatric DRE.

High frequency oscillations have been extensively investigated as interictal biomarkers of epilepsy. Yet, their value is largely debated due to the presence of physiological oscillations, which complicate distinguishing between normal versus abnormal events. So far, this debate has been addressed using intracranial EEG data from patients with drug-resistant epilepsy. Yet, this approach suffers from inability to record control data from healthy subjects and lack of whole brain coverage. Here, we aim to differentiate physiological from pathological high frequency oscillations using non-invasive whole brain electrophysiological recordings from children with drug-resistant epilepsy and typically developing controls. We recorded high-density EEG and magnetoencephalography data from 47 controls (median age: 11 years; 25 females) and 54 children with drug-resistant epilepsy (median age: 14 years, 33 females). We detected high frequency oscillations (in ripple frequency band) semi-automatically and localized their cortical generators through electric or magnetic source imaging. From each ripple, we extracted a set of temporal, morphological, spectral and spatial features. We then compared the features between ripples recorded from the epileptic brain (further distinguished into those from epileptogenic and non-epileptogenic regions) and those recorded from the control group (normal brain). We used these features to cross-validate a Naïve-Bayes algorithm for classifying each ripple recorded from children with epilepsy as coming from an epileptogenic region or not. We observed more high frequency oscillations on EEG than magnetoencephalography recordings (P < 0.001) both in the epilepsy and control groups. Physiological high frequency oscillations (recorded from controls) showed lower power, shorter duration and less variability (in both amplitude and duration) than those recorded from the epilepsy group (P < 0.001). Inter-channel latency of physiological ripples was longer compared to ripples from the epileptogenic regions (P < 0.01), while it was similar to the ripples from non-epileptogenic regions (P > 0.05). Ripples from epileptogenic regions showed larger extent than those from non-epileptogenic regions or from the control group (P < 0.001). The classification model showed an accuracy of 73%, with negative and positive predictive values of 73% and 70% (P < 0.0001), respectively, in classifying high frequency oscillations from the drug-resistant epilepsy group (as either epileptogenic or not). Our study indicates that physiological high frequency oscillations, recorded from the healthy brain, have distinct temporal, morphological, spectral and spatial features compared to those generated by the epileptic brain. The differentiation of pathological from physiological high frequency oscillations through non-invasive full-head techniques may augment the presurgical evaluation process of children with drug-resistant epilepsy and lead to better postsurgical seizure outcomes.

Spikes are the most established interictal epilepsy biomarkers. Yet, they suffer from low specificity since they are partially concordant with the epileptogenic zone and are often found in non-epileptogenic areas. High-frequency oscillations, classified as ripples and fast ripples, are considered more specific biomarkers compared with spikes. Ripples occur more often than fast ripples but are believed to be less specific, since they are more frequently generated by physiological mechanisms. Here, we examine the temporal relationship between spikes, ripples and fast ripples, and assess the ability of these biomarkers (and their combinations) to delineate the epileptogenic zone and predict outcome. We hypothesize that spikes on ripples (temporal co-occurrence of spikes and ripples) can identify the epileptogenic zone and predict outcome better than spikes or ripples. We analysed intracranial EEG data from 40 children with drug-resistant epilepsy. Spikes, ripples and fast ripples were classified based on their temporal occurrence. Their rates were compared with resection by performing a receiver operating characteristic analysis. The resection ratio, quantifying the extent of each biomarker's removal, was computed, and correlated with patients' outcome. Spikes on ripples were seen in all patients; fast ripples were seen in 43% of patients. In good outcome patients, fast ripple and spike on ripple rates were higher inside resection (P = 0.027; P = 0.003, respectively). Fast ripples and spikes on ripples resection ratio predicted outcome (P < 0.05). For fast ripples, outcome was predicted in 82% of patients; this proportion was higher than the one for spikes (48%, P = 0.015) and ripples (40%, P = 0.003), and spikes on ripples (53%, P = 0.034). Fast ripples were the most accurate (82%) to predict outcome; spikes on ripples were the most precise (positive predictive value = 90%). Spike rate and spikes on ripples performance to predict the epileptogenic zone were correlated (r = 0.36, P = 0.035). For patients with frequent spikes, spikes on ripples accuracy to predict outcome reached 70%. Fast ripples are the best biomarker, but they can be seen in only half of patients with drug-resistant epilepsy. Spikes on ripples are a good alternative with more universal applicability since they can be seen in all patients while their resection predicts good outcome; their performance is improved in patients with frequent spikes. Overall, in the absence of fast ripples, spike on ripple areas should be targeted during surgery.

Surgical success for patients with focal drug resistant epilepsy (DRE) relies on accurate localization of the epileptogenic zone (EZ). Currently, no exam delineates this zone unambiguously. Instead, the EZ is approximated by the area where seizures begin, which is identified manually through a tedious process that is prone to errors and biases. More importantly, resection of this area does not always predict good surgical outcome. Here, we propose an artificially intelligent, patient-specific framework that automatically identifies the EZ requiring little to no input from clinicians, without having to wait for a seizure to occur. The framework transforms interictal intracranial electroencephalography data into spatiotemporal representations of brain activity discriminating the interictal epileptogenic network from background activity. The epileptogenic network delineates the EZ with high precision and predicts surgical outcome. Our framework eliminates the need for manual data inspection, reduces prolonged monitoring, and enhances surgical planning for DRE patients.

In drug-resistant epilepsy, a visual inspection of intracranial electroencephalography (iEEG) signals is often needed to localize the epileptogenic zone (EZ) and guide neurosurgery. The visual assessment of iEEG time-frequency (TF) images is an alternative to signal inspection, but subtle variations may escape the human eye. Here, we propose a deep learning-based metric of visual complexity to interpret TF images extracted from iEEG data and aim to assess its ability to identify the EZ in the brain. We analyzed interictal iEEG data from 1928 contacts recorded from 20 children with drug-resistant epilepsy who became seizure-free after neurosurgery. We localized each iEEG contact in the MRI, created TF images (1-70 Hz) for each contact, and used a pre-trained VGG16 network to measure their visual complexity by extracting unsupervised activation energy (UAE) from 13 convolutional layers. We identified points of interest in the brain using the UAE values via patient- and layer-specific thresholds (based on extreme value distribution) and using a support vector machine classifier. Results show that contacts inside the seizure onset zone exhibit lower UAE than outside, with larger differences in deep layers (L10, L12, and L13: p < 0.001). Furthermore, the points of interest identified using the support vector machine, localized the EZ with 7 mm accuracy. In conclusion, we presented a pre-surgical computerized tool that facilitates the EZ localization in the patient's MRI without requiring long-term iEEG inspection.

OBJECTIVE: Interictal biomarkers are critical for identifying the epileptogenic focus. However, spikes and ripples lack specificity while fast ripples lack sensitivity. These biomarkers propagate from more epileptogenic onset to areas of spread. The pathophysiological mechanism of these propagations is elusive. Here, we examine zones where spikes and high frequency oscillations co-occur (SHFO), the spatiotemporal propagations of spikes, ripples, and fast ripples, and evaluate the spike-ripple onset overlap (SRO) as an epilepsy biomarker.

METHODS: We retrospectively analyzed intracranial EEG data from 41 patients with drug-resistant epilepsy. We mapped propagations of spikes, ripples, and fast ripples, and identified their onset and spread zones, as well as SHFO and SRO. We then estimated the SRO prognostic value in predicting surgical outcome and compared it to onset and spread zones of spike, ripple, and fast ripple propagations, and SHFO.

RESULTS: We detected spikes and ripples in all patients and fast ripples in 12 patients (29%). We observed spike and ripple propagations in 40 (98%) patients. Spike and ripple onsets overlapped in 35 (85%) patients. In good outcome patients, SRO showed higher specificity and precision (p < 0.05) in predicting resection compared to onset and zones of spikes, ripples, and SHFO. Only SRO resection predicted outcome (p = 0.01) with positive and negative predictive values of 82% and 57%, respectively.

INTERPRETATION: SRO is a specific and precise biomarker of the epileptogenic zone whose removal predicts outcome. SRO is present in most patients with drug-resistant epilepsy. Such a biomarker may reduce prolonged intracranial monitoring and improve outcome.