There is an increasing interest in applying artificial intelligence techniques to forecast epileptic seizures. In particular, machine learning algorithms could extract nonlinear statistical regularities from electroencephalographic (EEG) time series that can anticipate abnormal brain activity. The recent literature reports promising results in seizure detection and prediction tasks using machine and deep learning methods. However, performance evaluation is often based on questionable randomized cross-validation schemes, which can introduce correlated signals (e.g., EEG data recorded from the same patient during nearby periods of the day) into the partitioning of training and test sets. The present study demonstrates that the use of more stringent evaluation strategies, such as those based on leave-one-patient-out partitioning, leads to a drop in accuracy from about 80% to 50% for a standard eXtreme Gradient Boosting (XGBoost) classifier on two different data sets. Our findings suggest that the definition of rigorous evaluation protocols is crucial to ensure the generalizability of predictive models before proceeding to clinical trials.

No abstract

No abstract

In epileptic encephalopathies (EE), interictal epileptiform discharges (IEDs) contribute to cognitive impairment. The EE process has been studied in a patient affected by epilepsy with occipital calcification and celiac disease (CEC syndrome) by combining the administration of brain area stimulus specific (visual and auditory) reaction times (RT) during continuous EEG monitoring with the off-line reconstruction of auditory and visual evoked potentials (EP). Visual RT and VEP were abnormal only if recorded concomitantly to the IEDs. Auditory RT and EP were normal. When the EE process is going on, IEDs transiently disrupt aspects of cortical functioning, contributing to the cognitive impairment.

Temporal lobe epilepsy (TLE) is a brain network disorder characterized by alterations at both the structural and the functional level. It remains unclear how structure and function are related and whether this has any clinical relevance. In the present work, we adopted a novel methodological approach investigating how network structural features influence the large-scale dynamics. The functional network was defined by the spatio-temporal spreading of aperiodic bursts of activations (neuronal avalanches), as observed utilizing high-density electroencephalography (hdEEG) in TLE patients. The structural network was modeled as the region-based thickness covariance. Loosely speaking, we quantified the similarity of the cortical thickness of any two brain regions, both across groups, and at the individual level, the latter utilizing a novel approach to define the personalized covariance network (pCN). In order to compare the structural and functional networks (at the nodal level), we studied the correlation between the probability that a wave of activity would propagate from a source to a target region, and the similarity of the source region thickness as compared to other target brain regions. Building on the recent evidence that large-waves of activities pathologically spread through the epileptogenic network in TLE, also during resting state, we hypothesize that the structural cortical organization might influence such altered spatio-temporal dynamics. We observed a stable cluster of structure-function correlation in the bilateral limbic areas across subjects, highlighting group specific features for left, right and bilateral TLE. The involvement of contralateral areas was observed in unilateral TLE. We showed that in temporal lobe epilepsy alterations of structural and functional networks pair in the regions where seizures propagate and are linked to disease severity. In this study we leveraged on a well-defined model of neurological disease and pushed forward personalization approaches potentially useful in clinical practice. Finally, the methods developed here could be exploited to investigate the relationship between structure-function networks at subject level in other neurological conditions.