Combining meta and ensemble learning to classify EEG for seizure detection DOI Creative Commons
Mingze Liu, Jie Liu,

Mengna Xu

и другие.

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Март 28, 2025

Despite two decades of extensive research into electroencephalogram (EEG)-based automated seizure detection analysis, the persistent imbalance between and non-seizure categories remains a significant challenge. This study integrated meta-sampling with an ensemble classifier to address issue imbalanced classification existing in detection. In this framework, meta-sampler was employed autonomously acquire undersampling strategies from EEG data. During each iteration, interacted external environment on single occasion objective deriving optimal sampling strategy through interactive learning process. It anticipated that would be derived learning. And then soft Actor-Critic algorithm non-differentiable optimization associated meta-sampler. Consequently, framework adaptively selected training data, learned effective cascaded classifiers unbalanced epileptic Besides, time domain, nonlinear entropy-based features were extracted five frequency bands (δ, θ, α, β, γ) by Semi-JMI fed framework. The proposed system achieved sensitivity 92.58%, specificity 92.51%, accuracy 92.52% scalp dataset. On intracranial dataset, average sensitivity, specificity, 98.56%, 98.82%, 98.7%, respectively. experimental comparisons demonstrated outperformed other state-of-the-art methods, showed robustness face label corruption.

Язык: Английский

A comprehensive evaluation of interpretable artificial intelligence for epileptic seizure diagnosis using an electroencephalogram: A systematic review DOI Creative Commons
Daraje Kaba Gurmessa, Worku Jimma

Digital Health, Год журнала: 2025, Номер 11

Опубликована: Фев. 1, 2025

Background Epilepsy is a sensitive social and health issue that causes sudden death in epilepsy. Awake sleep electroencephalogram (EEG) first test confirms 80% of patients with confirmed Explainable artificial intelligence (XAI) for epileptic seizures (ESs) emerged to overcome drawbacks (AI) models like lack right explain, fairness, trustworthiness, an overwhelming paper was published. However, there reporting interpretable performance tradeoffs, stating the most AI applied, describing useful waveforms learned XAI models, documenting areas interest, identifying relationship between frequency bands Therefore, this systematic review aims comprehensively evaluate interpretability methods used ES monitoring using EEG. Methods This study followed PRISMA guidelines review. Advanced search queries were hardheaded into five reputable databases. Rayyan online platform used. The disagreement resolved through discussions. Results Twenty-three papers are included. A total 14 datasets 16,200 populations participated all included studies. CHB-MIT Dataset frequently (12 times). Minimizing number will increase accuracy reduce memory Interpretability trade-offs observed studies Discussion result implies further needed on multi-modal care recommendations, onset early warning minimize unexpected epilepsy damage. Optimizing ESs needs more investigation. Subjective matrices must be investigated very well before being by XAI. has no ethical considerations associated it. It been registered PROSPERO: registration number: CRD42023479926.

Язык: Английский

Процитировано

0
Combining meta and ensemble learning to classify EEG for seizure detection DOI Creative Commons
Mingze Liu, Jie Liu,

Mengna Xu

и другие.

Scientific Reports, Год журнала: 2025, Номер 15(1)

Опубликована: Март 28, 2025

Despite two decades of extensive research into electroencephalogram (EEG)-based automated seizure detection analysis, the persistent imbalance between and non-seizure categories remains a significant challenge. This study integrated meta-sampling with an ensemble classifier to address issue imbalanced classification existing in detection. In this framework, meta-sampler was employed autonomously acquire undersampling strategies from EEG data. During each iteration, interacted external environment on single occasion objective deriving optimal sampling strategy through interactive learning process. It anticipated that would be derived learning. And then soft Actor-Critic algorithm non-differentiable optimization associated meta-sampler. Consequently, framework adaptively selected training data, learned effective cascaded classifiers unbalanced epileptic Besides, time domain, nonlinear entropy-based features were extracted five frequency bands (δ, θ, α, β, γ) by Semi-JMI fed framework. The proposed system achieved sensitivity 92.58%, specificity 92.51%, accuracy 92.52% scalp dataset. On intracranial dataset, average sensitivity, specificity, 98.56%, 98.82%, 98.7%, respectively. experimental comparisons demonstrated outperformed other state-of-the-art methods, showed robustness face label corruption.

Язык: Английский

Процитировано

0