A neuromorphic physiological signal processing system based on VO2 memristor for next-generation human-machine interface

Nature Communications, Год журнала: 2023, Номер 14(1)

Опубликована: Июнь 21, 2023

Abstract Physiological signal processing plays a key role in next-generation human-machine interfaces as physiological signals provide rich cognition- and health-related information. However, the explosion of data presents challenges for traditional systems. Here, we propose highly efficient neuromorphic system based on VO 2 memristors. The volatile positive/negative symmetric threshold switching characteristics memristors are leveraged to construct sparse-spiking yet high-fidelity asynchronous spike encoder signals. Besides, dynamical behavior is utilized compact Leaky Integrate Fire (LIF) Adaptive-LIF (ALIF) neurons, which incorporated into decision-making Long short-term memory Spiking Neural Network. demonstrates superior computing capabilities, needing only small-sized LSNNs attain high accuracies 95.83% 99.79% arrhythmia classification epileptic seizure detection, respectively. This work highlights potential constructing systems promoting interfaces.

Язык: Английский

---

A hybrid CNN-Bi-LSTM model with feature fusion for accurate epilepsy seizure detection

BMC Medical Informatics and Decision Making, Год журнала: 2025, Номер 25(1)

Опубликована: Янв. 6, 2025

The diagnosis and treatment of epilepsy continue to face numerous challenges, highlighting the urgent need for development rapid, accurate, non-invasive methods seizure detection. In recent years, advancements in analysis electroencephalogram (EEG) signals have garnered widespread attention, particularly area recognition. A novel hybrid deep learning approach that combines feature fusion efficient detection is proposed this study. First, Discrete Wavelet Transform (DWT) applied perform a five-level decomposition raw EEG signals, from which time–frequency nonlinear features are extracted decomposed sub-bands. To eliminate redundant features, Support Vector Machine-Recursive Feature Elimination (SVM-RFE) employed select most distinctive fusion. Finally, states classified using Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-Bi-LSTM). method was rigorously validated on Bonn New Delhi datasets. binary classification tasks, both D-E group (Bonn dataset) Interictal-Ictal (New achieved 100% accuracy, sensitivity, specificity, precision, F1-score. three-class task A-D-E dataset, model performed excellently, achieving 96.19% 95.08% 97.34% 97.49% 96.18% addition, further larger more clinically relevant CHB-MIT average metrics 98.43% 97.84% 99.21% 99.14% an F1 score 98.39%. Compared existing literature, our outperformed several studies similar underscoring effectiveness advancement presented research. findings indicate demonstrates high level detecting seizures, crucial aspect managing epilepsy. By improving accuracy detection, has potential significantly enhance process diagnosing treating individuals affected by This could lead tailored plans, timely interventions, ultimately, better quality life patients.

Язык: Английский

---

An empirical wavelet transform-based approach for motion artifact removal in electroencephalogram signals

Decision Analytics Journal, Год журнала: 2024, Номер 10, С. 100420 - 100420

Опубликована: Фев. 10, 2024

Motion artifacts reduce the quality of information in electroencephalogram (EEG) signals. In this study, we have developed an effective approach to mitigate motion EEG signals by using empirical wavelet transform (EWT) technique. Firstly, decompose into narrowband called intrinsic mode functions (IMFs). These IMFs are further processed suppress artifacts. our first approach, principal component analysis (PCA) is employed noise from these decomposed IMFs. second with noisy components identified variance measure, which then removed obtain artifact-suppressed signal. Our experiments conducted on a publicly available Physionet dataset demonstrate effectiveness suppressing More importantly, IMF-variance-based has provided significantly better performance than EWT-PCA based approach. Also, IMF-variance computationally more efficient proposed achieved average signal ratio (ΔSNR) 28.26 dB and surpassed existing methods for artifact removal.

Язык: Английский

---

Machine Learning Enabled P300 Classifier for Autism Spectrum Disorder Using Adaptive Signal Decomposition

Brain Sciences, Год журнала: 2023, Номер 13(2), С. 315 - 315

Опубликована: Фев. 13, 2023

Joint attention skills deficiency in Autism spectrum disorder (ASD) hinders individuals from communicating effectively. The P300 Electroencephalogram (EEG) signal-based brain-computer interface (BCI) helps these neurorehabilitation training to overcome this deficiency. detection of the signal is more challenging ASD as it noisy, has less amplitude, and a higher latency than other individuals. This paper presents novel application variational mode decomposition (VMD) technique BCI system involving subjects for identification. EEG decomposed into five modes using VMD. Thirty linear non-linear time frequency domain features are extracted each mode. Synthetic minority oversampling data augmentation performed class imbalance problem chosen dataset. Then, comparative analysis three popular machine learning classifiers application. VMD's fifth with support vector (fine Gaussian kernel) classifier gave best performance parameters, namely accuracy, F1-score, area under curve, 91.12%, 91.18%, 96.6%, respectively. These results better when compared state-of-the-art methods.

Язык: Английский

---

End-to-end model for automatic seizure detection using supervised contrastive learning

Engineering Applications of Artificial Intelligence, Год журнала: 2024, Номер 133, С. 108665 - 108665

Опубликована: Май 28, 2024

Язык: Английский

---

Epileptic seizure detection using CHB-MIT dataset: The overlooked perspectives

Royal Society Open Science, Год журнала: 2024, Номер 11(5)

Опубликована: Май 1, 2024

Epilepsy is a life-threatening neurological condition. Manual detection of epileptic seizures (ES) laborious and burdensome. Machine learning techniques applied to electroencephalography (EEG) signals are widely used for automatic seizure detection. Some key factors worth considering the real-world applicability such systems: (i) continuous EEG data typically has higher class imbalance; (ii) variability across subjects present in physiological as EEG; (iii) event more practical than random segment Most prior studies failed address these crucial altogether In this study, we intend investigate generalized cross-subject system using from CHB-MIT dataset that considers all overlooked aspects. A 5-second non-overlapping window extract 92 features 22 channels; however, most significant 32 each channel experimentation. Seizure classification done Random Forest (RF) classifier detection, followed by post-processing method Adopting above-mentioned essential aspects, proposed achieved 72.63% 75.34% sensitivity subject-wise 5-fold leave-one-out analyses, respectively. This study presents scenario ES detectors furthers understanding systems.

Язык: Английский

---

Software advancements in automatic epilepsy diagnosis and seizure detection: 10-year review

Artificial Intelligence Review, Год журнала: 2024, Номер 57(7)

Опубликована: Июнь 21, 2024

Abstract Epilepsy is a chronic neurological disorder that may be diagnosed and monitored using routine diagnostic tests like Electroencephalography (EEG). However, manual introspection analysis of EEG signals presently difficult repetitive task even for experienced neuro-technologists with high false-positive rates inter- intra-rater reliability. Software advancements Artificial Intelligence (AI) algorithms have the potential to early detect predict abnormal patterns observed in signals. The present review focuses on systematically reporting software their implementation hardware systems automatic epilepsy diagnosis seizure detection past 10 years. Traditional, hybrid, end-to-end AI-based pipelines associated datasets been discussed. summarizes compares reported articles, datasets, patents through various subjective objective parameters this field. Latest demonstrate can reduce time by at least 50% without compromising accuracy or event detection. A significant rise software-based pipelines, deep learning architectures real-time analysis, granted has noticed since 2011. More than twenty-eight developed automatically diagnose epileptic from 2001 2023. Extensive explainability tools, cross-dataset generalizations, reproducibility ablation experiments further improve existing There need development standardized protocols data collection its AI pipeline robust, reliability-free, diagnosis.

Язык: Английский

---

A depression detection approach leveraging transfer learning with single-channel EEG

Journal of Neural Engineering, Год журнала: 2025, Номер 22(3), С. 036001 - 036001

Опубликована: Май 2, 2025

Abstract Objective. Major depressive disorder (MDD) is a widespread mental that affects health. Many methods combining electroencephalography (EEG) with machine learning or deep have been proposed to objectively distinguish between MDD and healthy individuals. However, most current detect depression based on multichannel EEG signals, which constrains its application in daily life. The context obtained can vary terms of study designs equipment settings, the available data limited, could also potentially lessen efficacy model differentiating subjects. To solve above challenges, detection leveraging transfer single-channel advanced. Approach. We utilized pretrained ResNet152V2 network flattening layer dense were appended. method feature extraction was applied, meaning all layers within frozen only parameters newly added adjustable during training. Given superiority neural networks image processing, temporal sequences signals are first converted into images, transforming problem signal categorization an classification task. Subsequently, cross-subject experimental strategy adopted for training performance evaluation. Main results. capable precisely (approaching 100% accuracy) identifying other individuals by employing samples from limited number Furthermore, exhibited superior across four publicly datasets, thereby demonstrating good adaptability response variations caused context. Significance. This research not highlights impressive potential techniques analysis but paves way innovative technical approaches facilitate early diagnosis associated disorders future.

Язык: Английский

---

Brain Epileptic Seizure Detection Using Joint CNN and Exhaustive Feature Selection With RNN-BLSTM Classifier

IEEE Access, Год журнала: 2023, Номер 11, С. 97990 - 98004

Опубликована: Янв. 1, 2023

Brain Epilepsy seizure is a critical disorder, which an uncontrolled burst of electrical activity brain. The early detection brain can save the life humans. electroencephalogram (EEG) signals may be used to automatically identify seizures, one most prominent solutions for this issue. However, conventional methods are failed classify effectively. So, work implemented Seizure-Detection-Network (BESD-Net) using deep learning, recurrent learning properties. Initially, dataset pre-processing performed, eliminates noise, unwanted data from EEG dataset. Then, based customized convolution neural network (CCNN) trained on pre-processed precise extraction disease correlated features. machine exhaustive random forest (ERF) feature selection optimize features CCNN features, highly with dependent In conclusion, (RNN) bi-directional long short-term memory (BLSTM) in order detect seizures chosen ERF Training and testing suggested methodology had made use CHB-MIT Scalp Database. aforementioned model has achieved values 98.36%, 97.54%, 97.91%, 98% 95.08% respectively precision, sensitivity, F1-Score, accuracy specificity. findings simulations demonstrate that BESD-Net led superior performance when compared technologies already use.

Язык: Английский

---

Classification of Epileptic Seizures Using LSTM Based Zebra Optimization Algorithm with Hyperparameter Tunin

International journal of intelligent engineering and systems, Год журнала: 2024, Номер 17(3), С. 80 - 91

Опубликована: Май 3, 2024

Electroencephalogram (EEG) are the neuro-electrophysiology signals, which commonly used as a diagnostic tool to measure seizure activity of brain.The accurate detection and classification seizures help provide an optimal solution diagnose patient.In this research, hyperparameter tuning with Zebra Optimization Algorithm (ZOA) is proposed for fine features from EEG signals.The signals taken three standard datasets such Temple University Hospital (TUH) at rate sampling signal 250Hz, Bonn (BU) 173.61Hz rate, Bern Barcelona (BB) alongside frequency 512 Hz.The pre-processed using Butterworth 8th order filtering method remove unwanted noise, de-noised decomposed by swarm decomposition method.Features like statistic-based features, frequency-dependent multi-scale wavelet transformation, entropy power spectral extracted then undergo ZOA followed feature selection Enhanced Spatial bound Whale (WOA) combination Salp Swarm (SSA) hybridized Lens Opposition-based Learning (LOBL) mechanism.The obtained algorithm fed hyper parameter optimized Long Short-Term Memory (LSTM) classifier classify normality abnormality seizures.The attained outcomes suggested approach exhibit better 98.43% accuracy on BU dataset, 99.71% BB TUH dataset.

Язык: Английский

---