Automatic EEG artifact detection using a local-global feature fusion network in time and time-frequency domains

Area of Science:

• Neurology
• Medical Imaging
• Artificial Intelligence

Background:

• Video electroencephalography (VEEG) interpretation is challenged by artifacts, impacting diagnostic accuracy and clinician efficiency.
• Artifacts from automatisms in VEEG can provide crucial diagnostic insights for temporal lobe epilepsy (TLE).

Purpose of the Study:

• To develop an automated algorithm for identifying and classifying VEEG artifacts.
• To aid clinicians in differentiating between interference signals and seizure-related activity in VEEG.

Main Methods:

• Proposed a Local-Global Feature Fusion Network based on Time-Domain and Time-Frequency Domain (LG-TDTFD-Net).
• Employed Convolutional Neural Networks (CNNs) and Transformer for extracting local and global features from VEEG signals.
• Integrated time-domain and time-frequency domain analysis for comprehensive feature extraction.

Main Results:

• The LG-TDTFD-Net demonstrated superior performance in detecting and classifying TLE artifacts.
• The proposed method outperformed existing baseline models on both clinical and public datasets.

Conclusions:

• The fusion of local and global features enhances VEEG artifact representation and model generalization.
• The developed algorithm effectively distinguishes artifacts from seizure signals, supporting clinical diagnosis of TLE.