Evaluation of an automated spike-and-wave complex detection algorithm in the EEG from a rat model of absence epilepsy

Automated recognition of epilepsy from EEG signals using a combining space-time algorithm of CNN-LSTM

Intelligent recognition methods for classifying non-stationary and non-invasive epileptic diagnoses are essential tools in neurological research. Electroencephalogram (EEG) signals exhibit better temporal characteristics in the detection of epilepsy compared to radiation medical images like computed tomography (CT) and magnetic resonance imaging (MRI), as they provide real-time insights into the disease' condition. While classical machine learning methods have been used for epilepsy EEG classification, they still often require manual parameter adjustments. Previous studies primarily focused on binary epilepsy recognition (epilepsy vs. healthy subjects) rather than as ternary status recognition (continuous epilepsy vs. intermittent epilepsy vs. healthy subjects). In this study, we propose a novel deep learning method that combines a convolution neural network (CNN) with a long short-term memory (LSTM) network for multi-class classification including both binary and ternary tasks, using a publicly available benchmark database on epilepsy EEGs. The hybrid CNN-LSTM automatically acquires knowledge without the need for extra pre-processing or manual intervention. Besides, the joint network method benefits from memory function and stronger feature extraction ability. Our proposed hybrid CNN-LSTM achieves state-of-the-art performance in ternary classification, outperforming classical machine learning and the latest deep learning models. For the three-class classification, in the method achieves an accuracy, specificity, sensitivity, and ROC of 98%, 97.4, 98.3% and 96.8%, respectively. In binary classification, the method achieves better results, with ACC of 100%, 100%, and 99.8%, respectively. Our dual stream spatiotemporal hybrid network demonstrates superior performance compared to other methods. Notably, it eliminates the need for manual operations, making it more efficient for doctors to diagnose during the clinical process and alleviating the workload of neurologists.

EEG phase synchronization during absence seizures

Absence seizures-generalized rhythmic spike-and-wave discharges (SWDs) are the defining property of childhood (CAE) and juvenile (JAE) absence epilepsies. Such seizures are the most compelling examples of pathological neuronal hypersynchrony. All the absence detection algorithms proposed so far have been derived from the properties of individual SWDs. In this work, we investigate EEG phase synchronization in patients with CAE/JAE and healthy subjects to explore the possibility of using the wavelet phase synchronization index to detect seizures and quantify their disorganization (fragmentation). The overlap of the ictal and interictal probability density functions was high enough to preclude effective seizure detection based solely on changes in EEG synchronization. We used a machine learning classifier with the phase synchronization index (calculated for 1 s data segments with 0.5 s overlap) and the normalized amplitude as features to detect generalized SWDs. Using 19 channels (10-20 setup), we identified 99.2% of absences. However, the overlap of the segments classified as ictal with seizures was only 83%. The analysis showed that seizures were disorganized in approximately half of the 65 subjects. On average, generalized SWDs lasted about 80% of the duration of abnormal EEG activity. The disruption of the ictal rhythm can manifest itself as the disappearance of epileptic spikes (with high-amplitude delta waves persisting), transient cessation of epileptic discharges, or loss of global synchronization. The detector can analyze a real-time data stream. Its performance is good for a six-channel setup (Fp1, Fp2, F7, F8, O1, O2), which can be implemented as an unobtrusive EEG headband. False detections are rare for controls and young adults (0.03% and 0.02%, respectively). In patients, they are more frequent (0.5%), but in approximately 82% cases, classification errors are caused by short epileptiform discharges. Most importantly, the proposed detector can be applied to parts of EEG with abnormal EEG activity to quantitatively determine seizure fragmentation. This property is important because a previous study reported that the probability of disorganized discharges is eight times higher in JAE than in CAE. Future research must establish whether seizure properties (frequency, length, fragmentation, etc.) and clinical characteristics can help distinguish CAE and JAE.

Absence Seizure Detection Algorithm for Portable EEG Devices

Absence seizures are generalized nonmotor epileptic seizures with abrupt onset and termination. Transient impairment of consciousness and spike-slow wave discharges (SWDs) in EEG are their characteristic manifestations. This type of seizure is severe in two common pediatric syndromes: childhood (CAE) and juvenile (JAE) absence epilepsy. The appearance of low-cost, portable EEG devices has paved the way for long-term, remote monitoring of CAE and JAE patients. The potential benefits of this kind of monitoring include facilitating diagnosis, personalized drug titration, and determining the duration of pharmacotherapy. Herein, we present a novel absence detection algorithm based on the properties of the complex Morlet continuous wavelet transform of SWDs. We used a dataset containing EEGs from 64 patients (37 h of recordings with almost 400 seizures) and 30 age and sex-matched controls (9 h of recordings) for development and testing. For seizures lasting longer than 2 s, the detector, which analyzed two bipolar EEG channels (Fp1-T3 and Fp2-T4), achieved a sensitivity of 97.6% with 0.7/h detection rate. In the patients, all false detections were associated with epileptiform discharges, which did not yield clinical manifestations. When the duration threshold was raised to 3 s, the false detection rate fell to 0.5/h. The overlap of automatically detected seizures with the actual seizures was equal to ~96%. For EEG recordings sampled at 250 Hz, the one-channel processing speed for midrange smartphones running Android 10 (about 0.2 s per 1 min of EEG) was high enough for real-time seizure detection.

An automated, machine learning-based detection algorithm for spike-wave discharges (SWDs) in a mouse model of absence epilepsy

OBJECTIVE

Manual detection of spike-wave discharges (SWDs) from electroencephalography (EEG) records is time intensive, costly, and subject to inconsistencies/biases. In addition, manual scoring often omits information on SWD confidence/intensity, which may be important for the investigation of mechanistic-based research questions. Our objective is to develop an automated method for the detection of SWDs in a mouse model of absence epilepsy that is focused on the characteristics of human scoring of preselected events to establish a confidence-based, continuous-valued scoring.

METHODS

We develop a support vector machine (SVM)-based algorithm for the automated detection of SWDs in the γ2R43Q mouse model of absence epilepsy. The algorithm first identifies putative SWD events using frequency- and amplitude-based peak detection. Four humans scored a set of 2500 putative events identified by the algorithm. Then, using predictors calculated from the wavelet transform of each event and the labels from human scoring, we trained an SVM to classify (SWD/nonSWD) and assign confidence scores to each event identified from 60, 24-hour EEG records. We provide a detailed assessment of intra- and interrater scoring that demonstrates advantages of automated scoring.

RESULTS

The algorithm scored SWDs along a continuum that is highly correlated with human confidence and that allows us to more effectively characterize ambiguous events. We demonstrate that events along our scoring continuum are temporally and proportionately correlated with abrupt changes in spectral power bands relevant to normal behavioral states including sleep.

SIGNIFICANCE

Although there are automated and semi-automated methods for the detection of SWDs in humans and rats, we contribute to the need for continued development of SWD detection in mice. Our results demonstrate the value of viewing detection of SWDs as a continuous classification problem to better understand "ground truth" in SWD detection (ie, the most reliable features agreed upon by humans that also correlate with objective physiologic measures).

Automated quantification of EEG spikes and spike clusters as a new read out in Theiler's virus mouse model of encephalitis-induced epilepsy

Intracerebral infection of C57BL/6 mice with Theiler's murine encephalomyelitis virus (TMEV) replicates many features of viral encephalitis-induced epilepsy in humans, including neuroinflammation, early (insult-associated) and late (spontaneous) seizures, neurodegeneration in the hippocampus, and cognitive and behavioral alterations. Thus, this model may be ideally suited to study mechanisms involved in encephalitis-induced epilepsy as potential targets for epilepsy prevention. However, spontaneous recurrent seizures (SRS) occur too infrequently to be useful as a biomarker of epilepsy, e.g., for drug studies. This prompted us to evaluate whether epileptiform spikes or spike clusters in the cortical electroencephalogram (EEG) may be a useful surrogate of epilepsy in this model. For this purpose, we developed an algorithm that allows efficient and large-scale EEG analysis of early and late seizures, spikes, and spike clusters in the EEG. While 77% of the infected mice exhibited early seizures, late seizures were only observed in 33% of the animals. The clinical characteristics of early and late seizures did not differ except that late generalized convulsive (stage 5) seizures were significantly longer than early stage 5 seizures. Furthermore, the frequency of SRS was much lower than the frequency of early seizures. Continuous (24/7) video-EEG monitoring over several months following infection indicated that the latent period to onset of SRS was 61 (range 16-91) days. Spike and spike clusters were significantly more frequent in infected mice with late seizures than in infected mice without seizures or in mock-infected sham controls. Based on the results of this study, increases in EEG spikes and spike clusters in groups of infected mice may be used as a new readout for studies on antiepileptogenic or disease-modifying drug effects in this model, because the significant increase in average spike counts in mice with late seizures obviously indicates a proepileptogenic alteration.

Evaluation of the Biocompatibility of Polypyrrole Implanted Subdurally in GAERS

This blinded controlled prospective randomized study investigates the biocompatibility of polypyrrole (PPy) polymer that will be used for intracranial triggered release of anti-epileptic drugs (AEDs). Three by three millimeters PPy are implanted subdurally in six adult female genetic absence epilepsy rats from Strasbourg. Each rat has a polymer implanted on one side of the cortex and a sham craniotomy performed on the other side. After a period of seven weeks, rats are euthanized and parallel series of coronal sections are cut throughout the implant site. Four series of 15 sections are histological (hematoxylin and eosin) and immunohistochemically (neuron-specific nuclear protein, glial fibrillary acidic protein, and anti-CD68 antibody) stained and evaluated by three investigators. The results show that implanted PPy mats do not induce obvious inflammation, trauma, gliosis, and neuronal toxicity. Therefore the authors conclude the PPy used offer good histocompatibility with central nervous system cells and that PPy sheets can be used as intracranial, AED delivery implant.

Antiepileptic Effects of Lacosamide Loaded Polymers Implanted Subdurally in GAERS

The current experiment investigated the ability of coaxial electrospun poly(D,L-lactide-co-glycolide) (PLGA) biodegradable polymer implants loaded with the antiepileptic drugs (AED) lacosamide to reduce seizures following implantation above the motor cortex in the Genetic Absence Epilepsy Rat from Strasbourg (GAERS). In this prospective, randomized, masked experiments, GAERS underwent surgery for implantation of skull electrodes (n=6), skull electrodes and blank polymers (n=6), or skull electrodes and lacosamide loaded polymers (n=6). Thirty-minute electroencephalogram (EEG) recordings were started at day 7 after surgery and continued for eight weeks. The number of SWDs and mean duration of one SWD were compared week-by-week between the three groups. There was no difference in the number of SWDs between any of the groups. However, the mean duration of one SWD was significantly lower in the lacosamide polymer group for up to 7 weeks when compared to the control group (0.004<p<0.038). The mean duration of one seizure was also lower at weeks 3, 5, 6, and 7 when compared to the blank polymer group (p= 0.016, 0.037, 0.025, and 0.025, resp.). We have demonstrated that AED loaded PLGA polymer sheets implanted on the surface of the cortex could affect seizure activity in GAERS for a sustained period.

Clonic Seizures in GAERS Rats after Oral Administration of Enrofloxacin

The aim of this study was to evaluate the effect of oral enrofloxacin on the epileptic status of Genetic Absence Epilepsy Rats from Strasbourg (GAERS). Five adult female GAERS rats, with implanted extradural electrodes for EEG monitoring, were declared free of clonic seizures after an 8-wk observation period. Enrofloxacin was then added to their drinking water (42.5 mg in 750 mL), and rats were observed for another 3 days. The number of spike-and-wave discharges and mean duration of a single discharge did not differ before and after treatment, but 2 of the 5 rats developed clonic seizures after treatment. Enrofloxacin should be used with caution in GAERS rats because it might induce clonic seizures.