Truace
Subscribe

A jointly optimized hybrid framework for epileptic EEG artifact classification

Accurate classification of artifacts in long-term electroencephalography (EEG) recordings is essential for reliable automated epilepsy analysis, as artifacts may obscure epileptiform patterns and degrade downstream detection performance. However, existing methods still face challenges in complementary feature integration and coordinated optimization of feature selection and classifier parameters. To address the limited feature representation capability of handcrafted methods and the limited interpretability of d…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-2141· cite & verify
A jointly optimized hybrid framework for epileptic EEG artifact classification
Use case analysis: the ambulatory EEG in navy medicine for traumatic brain injuries by Burner, Tahnyah D.. Public domain

In brief

Researchers built a framework that sorts artifacts in long-term EEG recordings from people with epilepsy. It fuses handcrafted temporal, spectral and entropy features with deep features from a lightweight ResNet-LSTM network. A new feature-selection method and an adaptive seagull optimizer then tune the selector and a Random Forest classifier together.

Artifacts can hide epileptiform patterns and weaken downstream seizure detection, so cleaner artifact labeling matters for automated epilepsy analysis. The results come from the Temple University Artifact Corpus and clinical recordings, and the abstract does not describe deployment in routine care or effects on patient outcomes. How well the method works in other hospitals or on other recording setups remains to be shown.

Main points

  1. Handcrafted temporal-spectral-entropy features are combined with deep features from a lightweight ResNet-LSTM network.
  2. A Class Neighborhood Discrimination method selects discriminative features from the fused set.
  3. An Adaptive Seagull Optimization Algorithm jointly tunes the feature selector and Random Forest classifier parameters.
  4. Accuracy was 99.30% on the originally constructed dataset and 98.15% on a recording-independent clinical test set.
  5. The authors report outperforming representative feature selection methods, optimization strategies and classifiers.

The gain

A jointly optimized hybrid classifier reached 99.30% accuracy on its constructed dataset and 98.15% on a separate recording-independent clinical test set, supporting more reliable automated epilepsy EEG analysis.

The rundown

The work targets two gaps the authors identify: handcrafted features have limited representational power, and deep learning approaches have limited interpretability. Combining both feature types is meant to address each weakness.

Rather than tuning feature selection and the classifier separately, the framework optimizes them in one unified process. This coordinated optimization is presented as a key difference from existing methods.

Evaluation drew on the Temple University Artifact Corpus plus clinical EEG recordings. Beyond accuracy, the authors report consistent performance on MCC, Kappa and GDR.

Sources

  1. Peer-reviewedComputers in Biology and Medicine2026-10-16

The debate