High performance clean versus artifact dry electrode EEG data classification using Convolutional Neural Network...

Area of Science:

• Neuroscience
• Machine Learning
• Biomedical Engineering

Background:

• Convolutional Neural Networks (CNNs) show promise for electroencephalography (EEG) artifact detection but require extensive datasets.
• Dry electrode EEG datasets are notably sparse, posing a challenge for developing robust artifact detection algorithms.
• Transfer learning offers a potential solution to overcome data scarcity in dry electrode EEG analysis.

Purpose of the Study:

• To develop a CNN-based algorithm for classifying clean versus artifact dry electrode EEG data.
• To leverage transfer learning to address the limited availability of dry electrode EEG datasets.
• To evaluate the performance of the developed algorithm in distinguishing artifactual from clean EEG signals.

Main Methods:

• Dry electrode EEG data were collected from 13 subjects with induced artifacts.
• A pre-trained CNN, initially trained on wet electrode EEG data, was fine-tuned using a sparse dry electrode dataset.
• A majority voting ensemble of three fine-tuned CNNs was employed for the final classification algorithm.

Main Results:

• The fine-tuned algorithm achieved a significantly improved test accuracy of 90.7%, compared to the pre-trained CNN's 65.6%.
• The classification algorithm demonstrated high performance with an F1-score of 90.2%, precision of 89.1%, and recall of 91.2%.
• The model was trained on 0.40 million and tested on 0.17 million overlapping EEG segments.

Conclusions:

• Transfer learning successfully enabled the development of a high-performing CNN algorithm for clean versus artifact classification in dry electrode EEG data.
• The study highlights the efficacy of transfer learning in overcoming data limitations for EEG artifact detection.
• The developed algorithm offers a promising solution for accurate artifact classification in sparse dry electrode EEG datasets.