Classification of high frequency oscillations in epileptic intracerebral EEG

Epileptic EEG patterns recognition through machine learning techniques and relevant time-frequency features

OBJECTIVES

The present study is designed to explore the process of epileptic patterns' automatic detection, specifically, epileptic spikes and high-frequency oscillations (HFOs), via a selection of machine learning (ML) techniques. The primary motivation for conducting such a research lies mainly in the need to investigate the long-term electroencephalography (EEG) recordings' visual examination process, often considered as a time-consuming and potentially error-prone procedure, requiring a great deal of mental focus and highly experimented neurologists. On attempting to resolve such a challenge, a number of state-of-the-art ML algorithms have been evaluated and compare in terms of performance, to pinpoint the most effective algorithm fit for accurately extracting epileptic EEG patterns.

CONTENT

Based on intracranial as well as simulated EEG data, the attained findings turn out to reveal that the randomforest (RF) method proved to be the most consistently effective approach, significantly outperforming the entirety of examined methods in terms of EEG recordings epileptic-pattern identification. Indeed, the RF classifier appeared to record an average balanced classification rate (BCR) of 92.38 % in regard to spikes recognition process, and 78.77 % in terms of HFOs detection.

SUMMARY

Compared to other approaches, our results provide valuable insights into the RF classifier's effectiveness as a powerful ML technique, fit for detecting EEG signals born epileptic bursts.

OUTLOOK

As a potential future work, we envisage to further validate and sustain our major reached findings through incorporating a larger EEG dataset. We also aim to explore the generative adversarial networks (GANs) application so as to generate synthetic EEG signals or combine signal generation techniques with deep learning approaches. Through this new vein of thought, we actually preconize to enhance and boost the automated detection methods' performance even more, thereby, noticeably enhancing the epileptic EEG pattern recognition area.

Epileptic seizure focus detection from interictal electroencephalogram: a survey

Electroencephalogram (EEG) is one of most effective clinical diagnosis modalities for the localization of epileptic focus. Most current AI solutions use this modality to analyze the EEG signals in an automated manner to identify the epileptic seizure focus. To develop AI system for identifying the epileptic focus, there are many recently-published AI solutions based on biomarkers or statistic features that utilize interictal EEGs. In this review, we survey these solutions and find that they can be divided into three main categories: (i) those that use of biomarkers in EEG signals, including high-frequency oscillation, phase-amplitude coupling, and interictal epileptiform discharges, (ii) others that utilize feature-extraction methods, and (iii) solutions based upon neural networks (an end-to-end approach). We provide a detailed description of seizure focus with clinical diagnosis methods, a summary of the public datasets that seek to reduce the research gap in epilepsy, recent novel performance evaluation criteria used to evaluate the AI systems, and guidelines on when and how to use them. This review also suggests a number of future research challenges that must be overcome in order to design more efficient computer-aided solutions to epilepsy focus detection.

Interictal high frequency background activity as a biomarker of epileptogenic tissue

High frequency oscillations (HFOs) are very brief events that are a well-established biomarker of the epileptogenic zone (EZ) but are rare and comprise only a tiny fraction of the total recorded EEG. We hypothesize that the interictal high frequency ‘background’ data, which has received little attention but represents the majority of the EEG record, also may contain additional, novel information for identifying the EZ. We analysed intracranial EEG (30–500 Hz frequency range) acquired from 24 patients who underwent resective surgery. We computed 38 quantitative features based on all usable, interictal data (63–307 h per subject), excluding all detected HFOs. We assessed association between each feature and the seizure onset zone (SOZ) and resected volume (RV) using logistic regression. A pathology score per channel was also created via principle component analysis and logistic regression, using hold-out-one-patient cross-validation to avoid in-sample training. Association of the pathology score with the SOZ and RV was quantified using an asymmetry measure. Many features were associated with the SOZ: 23/38 features had odds ratios >1.3 or <0.7 and 17/38 had odds ratios different than zero with high significance (P < 0.001/39, logistic regression with Bonferroni Correction). The pathology score, the rate of HFOs, and their channel-wise product were each strongly associated with the SOZ [median asymmetry ≥0.44, good surgery outcome patients; median asymmetry ≥0.40, patients with other outcomes; 95% confidence interval (CI) > 0.27 in both cases]. The pathology score and the channel-wise product also had higher asymmetry with respect to the SOZ than the HFO rate alone (median difference in asymmetry ≥0.18, 95% CI >0.05). These results support that the high frequency background data contains useful information for determining the EZ, distinct and complementary to information from detected HFOs. The concordance between the high frequency activity pathology score and the rate of HFOs appears to be a better biomarker of epileptic tissue than either measure alone.

Double-Step Machine Learning Based Procedure for HFOs Detection and Classification

The need for automatic detection and classification of high-frequency oscillations (HFOs) as biomarkers of the epileptogenic tissue is strongly felt in the clinical field. In this context, the employment of artificial intelligence methods could be the missing piece to achieve this goal. This work proposed a double-step procedure based on machine learning algorithms and tested it on an intracranial electroencephalogram (iEEG) dataset available online. The first step aimed to define the optimal length for signal segmentation, allowing for an optimal discrimination of segments with HFO relative to those without. In this case, binary classifiers have been tested on a set of energy features. The second step aimed to classify these segments into ripples, fast ripples and fast ripples occurring during ripples. Results suggest that LDA applied to 10 ms segmentation could provide the highest sensitivity (0.874) and 0.776 specificity for the discrimination of HFOs from no-HFO segments. Regarding the three-class classification, non-linear methods provided the highest values (around 90%) in terms of specificity and sensitivity, significantly different to the other three employed algorithms. Therefore, this machine-learning-based procedure could help clinicians to automatically reduce the quantity of irrelevant data.

High-Frequency Oscillations in the Scalp Electroencephalogram: Mission Impossible without Computational Intelligence

High-frequency oscillations (HFOs) in the electroencephalogram (EEG) are thought to be a promising marker for epileptogenicity. A number of automated detection algorithms have been developed for reliable analysis of invasively recorded HFOs. However, invasive recordings are not widely applicable since they bear risks and costs, and the harm of the surgical intervention of implantation needs to be weighted against the informational benefits of the invasive examination. In contrast, scalp EEG is widely available at low costs and does not bear any risks. However, the detection of HFOs on the scalp represents a challenge that was taken on so far mostly via visual detection. Visual detection of HFOs is, in turn, highly time-consuming and subjective. In this review, we discuss that automated detection algorithms for detection of HFOs on the scalp are highly warranted because the available algorithms were all developed for invasively recorded EEG and do not perform satisfactorily in scalp EEG because of the low signal-to-noise ratio and numerous artefacts as well as physiological activity that obscures the tiny phenomena in the high-frequency range.

Use of machine-learning algorithms to determine features of systolic blood pressure variability that predict poor outcomes in hypertensive patients

Background

We re-analyzed data from the Systolic Blood Pressure Intervention Trial (SPRINT) trial to identify features of systolic blood pressure (SBP) variability that portend poor cardiovascular outcomes using a nonlinear machine-learning algorithm.

Methods

We included all patients who completed 1 year of the study without reaching any primary endpoint during the first year, specifically: myocardial infarction, other acute coronary syndromes, stroke, heart failure or death from a cardiovascular event (n = 8799; 94%). In addition to clinical variables, features representing longitudinal SBP trends and variability were determined and combined in a random forest algorithm, optimized using cross-validation, using 70% of patients in the training set. Area under the curve (AUC) was measured using a 30% testing set. Finally, feature importance was determined by minimizing node impurity averaging over all trees in the forest for a specific feature.

Results

A total of 365 patients (4.1%) reached the combined primary outcome over 37 months of follow-up. The random forest classifier had an AUC of 0.71 on the testing set. The 10 most significant features selected in order of importance by the automated algorithm included the urine albumin/creatinine (CR) ratio, estimated glomerular filtration rate, age, serum CR, history of subclinical cardiovascular disease (CVD), cholesterol, a variable representing SBP signals using wavelet transformation, high-density lipoprotein, the 90th percentile of SBP and triglyceride level.

Conclusions

We successfully demonstrated use of random forest algorithm to define best prognostic longitudinal SBP representations. In addition to known risk factors for CVD, transformed variables for time series SBP measurements were found to be important in predicting poor cardiovascular outcomes and require further evaluation.

Computational modeling of high frequency oscillations recorded with clinical intracranial macroelectrodes

High Frequency Oscillations (HFOs) are a potential biomarker of epileptogenic regions. They have been extensively investigated in terms of automatic detection, classification and feature extraction. However, the mechanisms governing the generation of HFOs as well as the observability conditions on clinical intracranial macroelectrodes remain elusive. In this paper, we propose a novel physiologically-relevant macroscopic model for accurate simulation of HFOs as invasively recorded in epileptic patients. This model accounts for both the temporal and spatial properties of the cortical patch at the origin of epileptiform activity. Indeed, neuronal populations are combined with a 3D geometrical representation to simulate an extended epileptic source. Then, by solving the forward problem, the contributions of neuronal population signals are projected onto intracerebral electrode contacts. The obtained signals are qualitatively and quantitatively compared to real HFOs, and a relationship is drawn between macroscopic model parameters such as synchronization and spatial extent on the one hand, and HFO features such as the wave and fast ripple (200-600 Hz) components, on the other hand.