Intracerebral EEG Artifact Identification Using Convolutional Neural Networks.

Area of Science:

• Neuroscience
• Biomedical Engineering
• Machine Learning

Background:

• Manual identification of artifacts in intracerebral electroencephalographic (iEEG) recordings is time-consuming and prone to inaccuracies.
• Existing unsupervised methods lack the precision needed for reliable iEEG artifact detection.
• Development of automated, accurate artifact detection is crucial for large-scale iEEG data analysis.

Purpose of the Study:

• To introduce and evaluate a novel machine learning approach for detecting artifacts in iEEG signals.
• To benchmark the performance of the proposed method against expert annotations.
• To demonstrate the generalizability and adaptability of the developed model for diverse clinical settings.

Main Methods:

• A machine learning approach utilizing convolutional neural networks (CNN) was developed for iEEG artifact detection.
• The CNN model was trained and tested on iEEG data from St Anne's University Hospital and validated on data from Mayo Clinic.
• Transfer learning was explored for retraining the generalized model into a data-specific version.

Main Results:

• The proposed CNN technique demonstrated effective iEEG artifact detection, serving as a generalized model.
• The generalized and specialized models achieved F1 scores of 0.81 and 0.96 on the testing dataset, respectively.
• The CNN model significantly improved speed, objectivity, and reproducibility compared to manual artifact identification.

Conclusions:

• The developed CNN model provides a robust and efficient solution for automated iEEG artifact detection.
• The generalized model can be effectively retrained using transfer learning for specific EEG acquisition systems and noise environments.
• This approach enhances the reliability and efficiency of analyzing large intracerebral electroencephalographic datasets.