Transfer learning for multi-class artefact classification in dry EEG using convolutional neural networks
J.H.M. ter Braak (TU Delft - Mechanical Engineering)
Maritta van Stigt – Graduation committee member (Amsterdam UMC)
C. Strydis – Graduation committee member (Erasmus MC)
Mario Negrello – Graduation committee member (Erasmus MC)
Eva Groenendijk – Mentor (Amsterdam UMC)
More Info
expand_more
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons.
Abstract
Background Electroencephalography (EEG) using a dry electrode cap is currently being investigated as a pre-hospital stroke triage instrument. Developing an algorithm for automatic interpretation of the EEG signals is challenging, considering the amount of artefacts often in the signal. Ideally, an algorithm should be capable of distinguishing between different artefact types to determine the appropriate action: whether to correct them, reject them, or consider their potential predictive value. Neural networks have demonstrated their value for these types of classifications in wet EEG data. However, this approach requires enormous amount of data and dry EEG data is sparse.
Objective This study aims to develop a multi-class artefact classification model for dry EEG data using transfer learning.
Methods First, a convolutional neural network (CNN) for multi-class (clean, eye movement, muscle activity and electrode artefact) classification was developed. Wet electrode EEG recordings from a publicly available dataset were used, containing data of 213 patients (Part I). Second, this model was implemented and transfer learned for multi-class (clean, pulse artefact, muscle activity and artefact) classification of dry EEG recordings using data of 13 subjects (Part II). The models were trained using annotated multi-channel input. Model performances were evaluated on unseen test data using accuracy, area under the receiver operating characteristic curve, F1-score, precision, and recall.
Results The pre-trained multi-class model achieved an overall accuracy of 74.8%. The fine-tuned model was able to correctly differentiate between the classes with an accuracy of 71.2%, with the best performance for the classes muscle activity (AUC 0.92, F1-score 0.80) and artefact (AUC 0.94, F1-score 0.80).
Conclusion Transfer learning enabled the development of a good performing multi-class artefact classification model specifically tailored for dry EEG data, even though available data was limited. The developed model could assist in assessing appropriate action for different artefact types in dry EEG data interpretation algorithms.