eduzhai > Applied Sciences > Engineering >

Uncovering the structure of clinical EEG signals with self-supervised learning

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 32 pages

Abstract: Objective. Supervised learning paradigms are often limited by the amount oflabeled data that is available. This phenomenon is particularly problematic inclinically-relevant data, such as electroencephalography (EEG), where labelingcan be costly in terms of specialized expertise and human processing time.Consequently, deep learning architectures designed to learn on EEG data haveyielded relatively shallow models and performances at best similar to those oftraditional feature-based approaches. However, in most situations, unlabeleddata is available in abundance. By extracting information from this unlabeleddata, it might be possible to reach competitive performance with deep neuralnetworks despite limited access to labels. Approach. We investigatedself-supervised learning (SSL), a promising technique for discovering structurein unlabeled data, to learn representations of EEG signals. Specifically, weexplored two tasks based on temporal context prediction as well as contrastivepredictive coding on two clinically-relevant problems: EEG-based sleep stagingand pathology detection. We conducted experiments on two large public datasetswith thousands of recordings and performed baseline comparisons with purelysupervised and hand-engineered approaches. Main results. Linear classifierstrained on SSL-learned features consistently outperformed purely superviseddeep neural networks in low-labeled data regimes while reaching competitiveperformance when all labels were available. Additionally, the embeddingslearned with each method revealed clear latent structures related tophysiological and clinical phenomena, such as age effects. Significance. Wedemonstrate the benefit of self-supervised learning approaches on EEG data. Ourresults suggest that SSL may pave the way to a wider use of deep learningmodels on EEG data.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×