Epileptic Seizures Detection in EEG Signals Using Fusion Handcrafted and Deep Learning Features

Epileptic Seizure Detection with an End-to-End Temporal Convolutional Network and Bidirectional Long Short-Term Memory Model

Automatic seizure detection plays a key role in assisting clinicians for rapid diagnosis and treatment of epilepsy. In view of the parallelism of temporal convolutional network (TCN) and the capability of bidirectional long short-term memory (BiLSTM) in mining the long-range dependency of multi-channel time-series, we propose an automatic seizure detection method with a novel end-to-end TCN-BiLSTM model in this work. First, raw EEG is filtered with a 0.5-45 Hz band-pass filter, and the filtered data are input into the proposed TCN-BiLSTM network for feature extraction and classification. Post-processing process including moving average filtering, thresholding and collar technique is then employed to further improve the detection performance. The method was evaluated on two EEG database. On the CHB-MIT scalp EEG database, our method achieved a segment-based sensitivity of 94.31%, specificity of 97.13%, and accuracy of 97.09%. Meanwhile, an event-based sensitivity of 96.48% and an average false detection rate (FDR) of 0.38/h were obtained. On the SH-SDU database we collected, the segment-based sensitivity of 94.99%, specificity of 93.25%, and accuracy of 93.27% were achieved. In addition, an event-based sensitivity of 99.35% and a false detection rate of 0.54/h were yielded. The total detection time consumed for 1[Formula: see text]h EEG data was 5.65[Formula: see text]s. These results demonstrate the superiority and promising potential of the proposed method in real-time monitoring of epileptic seizures.

Exploring Convolutional Neural Network Architectures for EEG Feature Extraction

The main purpose of this paper is to provide information on how to create a convolutional neural network (CNN) for extracting features from EEG signals. Our task was to understand the primary aspects of creating and fine-tuning CNNs for various application scenarios. We considered the characteristics of EEG signals, coupled with an exploration of various signal processing and data preparation techniques. These techniques include noise reduction, filtering, encoding, decoding, and dimension reduction, among others. In addition, we conduct an in-depth analysis of well-known CNN architectures, categorizing them into four distinct groups: standard implementation, recurrent convolutional, decoder architecture, and combined architecture. This paper further offers a comprehensive evaluation of these architectures, covering accuracy metrics, hyperparameters, and an appendix that contains a table outlining the parameters of commonly used CNN architectures for feature extraction from EEG signals.

Fractal Geometry Meets Computational Intelligence: Future Perspectives

Characterizations in terms of fractals are typically employed for systems with complex and multiscale descriptions. A prominent example of such systems is provided by the human brain, which can be idealized as a complex dynamical system made of many interacting subunits. The human brain can be modeled in terms of observable variables together with their spatio-temporal-functional relations. Computational intelligence is a research field bridging many nature-inspired computational methods, such as artificial neural networks, fuzzy systems, and evolutionary and swarm intelligence optimization techniques. Typical problems faced by means of computational intelligence methods include those of recognition, such as classification and prediction. Although historically conceived to operate in some vector space, such methods have been recently extended to the so-called nongeometric spaces, considering labeled graphs as the most general example of such patterns. Here, we suggest that fractal analysis and computational intelligence methods can be exploited together in neuroscience research. Fractal characterizations can be used to (i) assess scale-invariant properties and (ii) offer numeric, feature-based representations to complement the usually more complex pattern structures encountered in neurosciences. Computational intelligence methods could be used to exploit such fractal characterizations, considering also the possibility to perform data-driven analysis of nongeometric input spaces, therby overcoming the intrinsic limits related to Euclidean geometry.

Robust Epileptic Seizure Detection Using Long Short-Term Memory and Feature Fusion of Compressed Time-Frequency EEG Images

Epilepsy is a prevalent neurological disorder with considerable risks, including physical impairment and irreversible brain damage from seizures. Given these challenges, the urgency for prompt and accurate seizure detection cannot be overstated. Traditionally, experts have relied on manual EEG signal analyses for seizure detection, which is labor-intensive and prone to human error. Recognizing this limitation, the rise in deep learning methods has been heralded as a promising avenue, offering more refined diagnostic precision. On the other hand, the prevailing challenge in many models is their constrained emphasis on specific domains, potentially diminishing their robustness and precision in complex real-world environments. This paper presents a novel model that seamlessly integrates the salient features from the time-frequency domain along with pivotal statistical attributes derived from EEG signals. This fusion process involves the integration of essential statistics, including the mean, median, and variance, combined with the rich data from compressed time-frequency (CWT) images processed using autoencoders. This multidimensional feature set provides a robust foundation for subsequent analytic steps. A long short-term memory (LSTM) network, meticulously optimized for the renowned Bonn Epilepsy dataset, was used to enhance the capability of the proposed model. Preliminary evaluations underscore the prowess of the proposed model: a remarkable 100% accuracy in most of the binary classifications, exceeding 95% accuracy in three-class and four-class challenges, and a commendable rate, exceeding 93.5% for the five-class classification.

Brain-Computer Interface: The HOL-SSA Decomposition and Two-Phase Classification on the HGD EEG Data

An efficient processing approach is essential for increasing identification accuracy since the electroencephalogram (EEG) signals produced by the Brain-Computer Interface (BCI) apparatus are nonlinear, nonstationary, and time-varying. The interpretation of scalp EEG recordings can be hampered by nonbrain contributions to electroencephalographic (EEG) signals, referred to as artifacts. Common disturbances in the capture of EEG signals include electrooculogram (EOG), electrocardiogram (ECG), electromyogram (EMG) and other artifacts, which have a significant impact on the extraction of meaningful information. This study suggests integrating the Singular Spectrum Analysis (SSA) and Independent Component Analysis (ICA) methods to preprocess the EEG data. The key objective of our research was to employ Higher-Order Linear-Moment-based SSA (HOL-SSA) to decompose EEG signals into multivariate components, followed by extracting source signals using Online Recursive ICA (ORICA). This approach effectively improves artifact rejection. Experimental results using the motor imagery High-Gamma Dataset validate our method's ability to identify and remove artifacts such as EOG, ECG, and EMG from EEG data, while preserving essential brain activity.

A Shallow Autoencoder Framework for Epileptic Seizure Detection in EEG Signals

This paper presents a trainable hybrid approach involving a shallow autoencoder (AE) and a conventional classifier for epileptic seizure detection. The signal segments of a channel of electroencephalogram (EEG) (EEG epochs) are classified as epileptic and non-epileptic by employing its encoded AE representation as a feature vector. Analysis on a single channel-basis and the low computational complexity of the algorithm allow its use in body sensor networks and wearable devices using one or few EEG channels for wearing comfort. This enables the extended diagnosis and monitoring of epileptic patients at home. The encoded representation of EEG signal segments is obtained based on training the shallow AE to minimize the signal reconstruction error. Extensive experimentation with classifiers has led us to propose two versions of our hybrid method: (a) one yielding the best classification performance compared to the reported methods using the k-nearest neighbor (kNN) classifier and (b) the second with a hardware-friendly architecture and yet with the best classification performance compared to other reported methods in this category using a support-vector machine (SVM) classifier. The algorithm is evaluated on the Children's Hospital Boston, Massachusetts Institute of Technology (CHB-MIT), and University of Bonn EEG datasets. The proposed method achieves 98.85% accuracy, 99.29% sensitivity, and 98.86% specificity on the CHB-MIT dataset using the kNN classifier. The best figures using the SVM classifier for accuracy, sensitivity, and specificity are 99.19%, 96.10%, and 99.19%, respectively. Our experiments establish the superiority of using an AE approach with a shallow architecture to generate a low-dimensionality yet effective EEG signal representation capable of high-performance abnormal seizure activity detection at a single-channel EEG level and with a fine granularity of 1 s EEG epochs.

Leveraging Multiple Distinct EEG Training Sessions for Improvement of Spectral-Based Biometric Verification Results

Most studies on EEG-based biometry recognition report results based on signal databases, with a limited number of recorded EEG sessions using the same single EEG recording for both training and testing a proposed model. However, the EEG signal is highly vulnerable to interferences, electrode placement, and temporary conditions, which can lead to overestimated assessments of the considered methods. Our study examined how different numbers of distinct recording sessions used as training sessions would affect EEG-based verification. We analyzed the original data from 29 participants with 20 distinct recorded sessions each, as well as 23 additional impostors with only one session each. We applied raw coefficients of power spectral density estimate, and the coefficients of power spectral density estimate converted to the decibel scale, as the input to a shallow neural network. Our study showed that the variance introduced by multiple recording sessions affects sensitivity. We also showed that increasing the number of sessions above eight did not improve the results under our conditions. For 15 training sessions, the achieved accuracy was 96.7 ± 4.2%, and for eight training sessions and 12 test sessions, it was 94.9 ± 4.6%. For 15 training sessions, the rate of successful impostor attacks over all attack attempts was 3.1 ± 2.2%, but this number was not significantly different from using six recording sessions for training. Our findings indicate the need to include data from multiple recording sessions in EEG-based recognition for training, and that increasing the number of test sessions did not significantly affect the obtained results. Although the presented results are for the resting-state, they may serve as a baseline for other paradigms.

An efficient deep learning framework for P300 evoked related potential detection in EEG signal

BACKGROUND

Incorporating the time-frequency localization properties of Gabor transform (GT), the complexity understandings of convolutional neural network (CNN), and histogram of oriented gradients (HOG) efficacy in distinguishing positive peaks can exhibit their characteristics to reveal an effective solution in the detection of P300 evoked related potential (ERP). By applying a drastic number of convolutional layers, the majority of deep networks elicit sufficient properties for the output determination, leading to gigantic and time-consuming structures. In this paper, we propose a novel deep learning framework by the combination of tuned GT, and modified HOG with the CNN as "TGT-MHOG-CNN" for detection of P300 ERP in EEG signal.

METHOD

In the proposed method, GT is tuned based on triangular function for EEG signals, and then spectrograms including time-frequency information are captured. The function's parameters are justified to differentiate the signals with the P300 component. Furthermore, HOG is modified (MHOG) for the 2-D EEG signal, and consequently, gradients patterns are extracted for the target potentials. MHOG is potent in distinguishing the positive peak in the general waveform; however, GT unravels time-frequency information, which is ignored in the gradient histogram. These outputs of GT and MHOG do not share the same nature in the images nor overlap. Therefore, more extensive information is reached without redundancy or excessive information by fusing them. Combining GT and MHOG provides different patterns which benefit CNN for more precise detection. Consequently, TGT-MHOG-CNN ends in a more straightforward structure than other networks, and therefore, the whole performance is acceptable with faster rates and very high accuracy.

RESULTS

BCI Competition II and III datasets are used to evaluate the performance of the proposed method. These datasets include a complete record for P300 ERP with BCI2000 using a paradigm, and it has numerous noises, including power and muscle-based noises. The objective is to predict the correct character in each provided character selection epochs. Compared to state-of-the-art methods, simulation results indicate striking abilities of the proposed framework for P300 ERP detection. Our best record reached the P300 ERP classification rates of over 98.7% accuracy and 98.7% precision for BCI Competition II and 99% accuracy and 100% precision for BCI Competition III datasets, with superiority in execution time for the mentioned datasets.

M1M2: Deep-Learning-Based Real-Time Emotion Recognition from Neural Activity

Emotion recognition, or the ability of computers to interpret people's emotional states, is a very active research area with vast applications to improve people's lives. However, most image-based emotion recognition techniques are flawed, as humans can intentionally hide their emotions by changing facial expressions. Consequently, brain signals are being used to detect human emotions with improved accuracy, but most proposed systems demonstrate poor performance as EEG signals are difficult to classify using standard machine learning and deep learning techniques. This paper proposes two convolutional neural network (CNN) models (M1: heavily parameterized CNN model and M2: lightly parameterized CNN model) coupled with elegant feature extraction methods for effective recognition. In this study, the most popular EEG benchmark dataset, the DEAP, is utilized with two of its labels, valence, and arousal, for binary classification. We use Fast Fourier Transformation to extract the frequency domain features, convolutional layers for deep features, and complementary features to represent the dataset. The M1 and M2 CNN models achieve nearly perfect accuracy of 99.89% and 99.22%, respectively, which outperform every previous state-of-the-art model. We empirically demonstrate that the M2 model requires only 2 seconds of EEG signal for 99.22% accuracy, and it can achieve over 96% accuracy with only 125 milliseconds of EEG data for valence classification. Moreover, the proposed M2 model achieves 96.8% accuracy on valence using only 10% of the training dataset, demonstrating our proposed system's effectiveness. Documented implementation codes for every experiment are published for reproducibility.

A Classification Model of EEG Signals Based on RNN-LSTM for Diagnosing Focal and Generalized Epilepsy

Epilepsy is a chronic neurological disorder caused by abnormal neuronal activity that is diagnosed visually by analyzing electroencephalography (EEG) signals.

BACKGROUND

Surgical operations are the only option for epilepsy treatment when patients are refractory to treatment, which highlights the role of classifying focal and generalized epilepsy syndrome. Therefore, developing a model to be used for diagnosing focal and generalized epilepsy automatically is important.

METHODS

A classification model based on longitudinal bipolar montage (LB), discrete wavelet transform (DWT), feature extraction techniques, and statistical analysis in feature selection for RNN combined with long short-term memory (LSTM) is proposed in this work for identifying epilepsy. Initially, normal and epileptic LB channels were decomposed into three levels, and 15 various features were extracted. The selected features were extracted from each segment of the signals and fed into LSTM for the classification approach.

RESULTS

The proposed algorithm achieved a 96.1% accuracy, a 96.8% sensitivity, and a 97.4% specificity in distinguishing normal subjects from subjects with epilepsy. This optimal model was used to analyze the channels of subjects with focal and generalized epilepsy for diagnosing purposes, relying on statistical parameters.

CONCLUSIONS

The proposed approach is promising, as it can be used to detect epilepsy with satisfactory classification performance and diagnose focal and generalized epilepsy.

Epileptic Seizure Detection Based on Variational Mode Decomposition and Deep Forest Using EEG Signals

Electroencephalography (EEG) records the electrical activity of the brain, which is an important tool for the automatic detection of epileptic seizures. It is certainly a very heavy burden to only recognize EEG epilepsy manually, so the method of computer-assisted treatment is of great importance. This paper presents a seizure detection algorithm based on variational modal decomposition (VMD) and a deep forest (DF) model. Variational modal decomposition is performed on EEG recordings, and the first three variational modal functions (VMFs) are selected to construct the time–frequency distribution of the EEG signals. Then, the log−Euclidean covariance matrix (LECM) is computed to represent the EEG properties and form EEG features. The deep forest model is applied to complete the EEG signal classification, which is a non-neural network deep model with a cascade structure that performs feature learning through the forest. In addition, to improve the classification accuracy, postprocessing techniques are performed to generate the discriminant results by moving average filtering and adaptive collar expansion. The algorithm was evaluated on the Bonn EEG dataset and the Freiburg long−term EEG dataset, and the former achieved a sensitivity and specificity of 99.32% and 99.31%, respectively. The mean sensitivity and specificity of this method for the 21 patients in the Freiburg dataset were 95.2% and 98.56%, respectively, with a false detection rate of 0.36/h. These results demonstrate the superior performance advantage of our algorithm and indicate its great research potential in epilepsy detection.

Impact of EEG Frequency Bands and Data Separation on the Performance of Person Verification Employing Neural Networks

The paper is devoted to the study of EEG-based people verification. Analyzed solutions employed shallow artificial neural networks using spectral EEG features as input representation. We investigated the impact of the features derived from different frequency bands and their combination on verification results. Moreover, we studied the influence of a number of hidden neurons in a neural network. The datasets used in the analysis consisted of signals recorded during resting state from 29 healthy adult participants performed on different days, 20 EEG sessions for each of the participants. We presented two different scenarios of training and testing processes. In the first scenario, we used different parts of each recording session to create the training and testing datasets, and in the second one, training and testing datasets originated from different recording sessions. Among single frequency bands, the best outcomes were obtained for the beta frequency band (mean accuracy of 91 and 89% for the first and second scenarios, respectively). Adding the spectral features from more frequency bands to the beta band features improved results (95.7 and 93.1%). The findings showed that there is not enough evidence that the results are different between networks using different numbers of hidden neurons. Additionally, we included results for the attack of 23 external impostors whose recordings were not used earlier in training or testing the neural network in both scenarios. Another significant finding of our study shows worse sensitivity results in the second scenario. This outcome indicates that most of the studies presenting verification or identification results based on the first scenario (dominating in the current literature) are overestimated when it comes to practical applications.

Classification of EEG Signals for Prediction of Epileptic Seizures

Epilepsy is a common brain disorder that causes patients to face multiple seizures in a single day. Around 65 million people are affected by epilepsy worldwide. Patients with focal epilepsy can be treated with surgery, whereas generalized epileptic seizures can be managed with medications. It has been noted that in more than 30% of cases, these medications fail to control epileptic seizures, resulting in accidents and limiting the patient’s life. Predicting epileptic seizures in such patients prior to the commencement of an oncoming seizure is critical so that the seizure can be treated with preventive medicines before it occurs. Electroencephalogram (EEG) signals of patients recorded to observe brain electrical activity during a seizure can be quite helpful in predicting seizures. Researchers have proposed methods that use machine and/or deep learning techniques to predict epileptic seizures using scalp EEG signals; however, prediction of seizures with increased accuracy is still a challenge. Therefore, we propose a three-step approach. It includes preprocessing of scalp EEG signals with PREP pipeline, which is a more sophisticated alternative to basic notch filtering. This method uses a regression-based technique to further enhance the SNR, with a combination of handcrafted, i.e., statistical features such as temporal mean, variance, and skewness, and automated features using CNN, followed by classification of interictal state and preictal state segments using LSTM to predict seizures. We train and validate our proposed technique on the CHB-MIT scalp EEG dataset and achieve accuracy of 94%, sensitivity of 93.8%, and 91.2% specificity. The proposed technique achieves better sensitivity and specificity than existing methods.

Engineering nonlinear epileptic biomarkers using deep learning and Benford's law

In this study, we designed two deep neural networks to encode 16 features for early seizure detection in intracranial EEG and compared them and their frequency responses to 16 widely used engineered metrics to interpret their properties: epileptogenicity index (EI), phase locked high gamma (PLHG), time and frequency domain Cho Gaines distance (TDCG, FDCG), relative band powers, and log absolute band powers (from alpha, beta, theta, delta, low gamma, and high gamma bands). The deep learning models were pretrained for seizure identification on the time and frequency domains of 1 s, single-channel clips of 127 seizures (from 25 different subjects) using "leave-one-out" (LOO) cross validation. Each neural network extracted unique feature spaces that were interpreted using spectral power modulations before being used to train a Random Forest Classifier (RFC) for seizure identification. The Gini Importance of each feature was calculated from the pretrained RFC, enabling the most significant features (MSFs) for each task to be identified. The MSFs were extracted to train another RFC for UPenn and Mayo Clinic's Seizure Detection Kaggle Challenge. They obtained an AUC score of 0.93, demonstrating a transferable method to identify and interpret biomarkers for seizure detection.