TATPat based explainable EEG model for neonatal seizure detection

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: Nov. 4, 2024

The most cost-effective data collection method is electroencephalography (EEG) to obtain meaningful information about the brain. Therefore, EEG signal processing very important for neuroscience and machine learning (ML). primary objective of this research detect neonatal seizures explain these using new version Directed Lobish. This uses a publicly available dataset get comparative results. In order classify signals, an explainable feature engineering (EFE) model has been proposed. EFE model, there are four essential phases phases: (i) automaton transformer-based extraction, (ii) selection deploying cumulative weight-based neighborhood component analysis (CWNCA), (iii) Lobish (DLob) Causal Connectome Theory (CCT)-based result generation (iv) classification t algorithm-based support vector (tSVM). first phase, we have used channel transformer numbers values divided into three levels named (1) high, (2) medium (3) low. By utilizing levels, created nodes (each node defines each level). extraction transition tables extracted. proposed function termed Triple Nodes Automaton-based Transition table Pattern (TATPat). contains 19 channels 9 (= 32) connection in defined automaton. Thus, presented TATPat extracts 3249 × 9) features from segment. To choose informative features, selector which CWNCA applied. cooperating findings DLob, results obtained. last phase high performance ensemble classifier (tSVM) obtained two validation techniques 10-fold cross-validation (CV) leave-one subject-out (LOSO) CV. generates DLob string by string, Moreover, attained 99.15% 76.37% accuracy LOSO CVs respectively. According performances, recommended TATPat-based good at classification. Also, artificial intelligence (XAI) since TTPat-based DLob.

Language: Английский

---

Electroencephalography Decoding with Conditional Identification Generator

International Journal of Neural Systems, Journal Year: 2025, Volume and Issue: unknown

Published: March 1, 2025

Decoding Electroencephalography (EEG) signals are extremely useful for advancing and understanding human–artificial intelligence (AI) interaction systems. Recent advancements in deep neural networks (DNNs) have demonstrated significant promise this respect due to their ability model complex nonlinear relationships. However, DNNs face persistent challenges addressing the inter-person variability inherent EEG signals, which limits generalizability. To tackle limitation, we propose a novel framework that integrates conditional identification information, leveraging between individual traits enhance model’s internal representation improve decoding accuracy. Building on foundation, further introduce privacy-preserving information generator — generative derives embedding knowledge directly from raw signals. This approach eliminates need personal via tests, ensuring both efficiency privacy. Experimental evaluations conducted WithMe dataset confirm outperforms baseline network architectures. Notably, our achieves substantial improvements accuracy familiar unseen subjects, paving way efficient, robust, privacy-conscious human–computer interface

Language: Английский

---

BPSSL: Balanced pseudo-label based semi-supervised learning for medical image classification

Biomedical Signal Processing and Control, Journal Year: 2025, Volume and Issue: 109, P. 108044 - 108044

Published: May 22, 2025

Language: Английский

---

Multiclass classification of epileptic seizure phases using a novel HFO-based feature extraction model

Signal Image and Video Processing, Journal Year: 2025, Volume and Issue: 19(4)

Published: Feb. 22, 2025

Language: Английский

---

An ensemble of fuzzy soft expert set with deep learning on attack detection for secure industrial cyber-physical systems

Journal of Radiation Research and Applied Sciences, Journal Year: 2025, Volume and Issue: 18(2), P. 101464 - 101464

Published: April 4, 2025

Language: Английский

---

LMA-EEGNet: A Lightweight Multi-Attention Network for Neonatal Seizure Detection Using EEG signals

Electronics, Journal Year: 2024, Volume and Issue: 13(12), P. 2354 - 2354

Published: June 16, 2024

Neonatal epilepsy is an early postnatal brain disorder, and automatic seizure detection crucial for timely diagnosis treatment to reduce potential damage. This work proposes a novel Lightweight Multi-Attention Network, LMA-EEGNet, diagnosing neonatal epileptic seizures from multi-channel EEG signals employing dilated depthwise separable convolution (DDS Conv) feature extraction using pointwise followed by global average pooling classification. The proposed approach substantially reduces the model size, number of parameters, computational complexity, which are real-time clinical seizures. LMA-EEGNet integrates temporal spectral features through distinct branches. branch uses DDS Conv extract features, enhanced channel attention mechanism. utilizes similar convolutions alongside spatial mechanism highlight key frequency components. Outputs both branches merged processed layer efficient detection. Experimental results show that our model, with only 2471 parameters size 23 KB, achieves accuracy 95.71% AUC 0.9862, demonstrating its practical deployment. study provides effective deep learning solution seizures, improving diagnostic timeliness.

Language: Английский

---

Efficient EEG feature learning model combining random convolutional kernel with wavelet scattering for seizure detection

International Journal of Neural Systems, Journal Year: 2024, Volume and Issue: 34(11)

Published: July 19, 2024

Automatic seizure detection has significant value in epilepsy diagnosis and treatment. Although a variety of deep learning models have been proposed to automatically learn electroencephalography (EEG) features for detection, the generalization performance computational burden such remain bottleneck practical application. In this study, novel lightweight model based on random convolutional kernel transform (ROCKET) is developed EEG feature detection. Specifically, kernels are embedded into structure wavelet scattering network instead original convolutions. Then selected from coefficients outputs by analysis variance (ANOVA) minimum redundancy-maximum relevance (MRMR) methods. This not only preserves merits fast-training process ROCKET, but also provides insight retaining helpful channels. The extreme gradient boosting (XGboost) classifier was combined with build comprehensive system that achieved promising epoch-based results, over 90% both sensitivity specificity scalp intracranial databases. experimental comparisons showed method outperformed other state-of-the-art methods cross-patient patient-specific

Language: Английский

---

Sparse Spike Feature Learning to Recognize Traceable Interictal Epileptiform Spikes

International Journal of Neural Systems, Journal Year: 2024, Volume and Issue: 35(02)

Published: Oct. 25, 2024

Interictal epileptiform spikes (spikes) and epileptogenic focus are strongly correlated. However, partial insensitive to focus, which restricts epilepsy neurosurgery. Therefore, identifying spike subtypes that associated with (traceable spikes) could facilitate their use as reliable signal sources for accurately tracing focus. the sparse firing phenomenon in transmission of intracranial neuronal discharges leads differences within cannot be observed visually. neuro-electro-physiologists unable identify traceable locate Herein, we propose a novel feature learning method recognize extract discrimination information related First, multilevel eigensystem representation was determined based on module express intrinsic properties spike. Second, expressed multi-domain context representations. Among them, encoding strategy implemented effectively simulate accurate activity neurosources. The sensitivity proposed 97.1%, demonstrating its effectiveness significant efficiency relative other state-of-the-art methods.

Language: Английский

---