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

OBJECTIVE

Supervised learning paradigms are often limited by the amount of labeled data that is available. This phenomenon is particularly problematic in clinically-relevant data, such as electroencephalography (EEG), where labeling can be costly in terms of specialized expertise and human processing time. Consequently, deep learning architectures designed to learn on EEG data have yielded relatively shallow models and performances at best similar to those of traditional feature-based approaches. However, in most situations, unlabeled data is available in abundance. By extracting information from this unlabeled data, it might be possible to reach competitive performance with deep neural networks despite limited access to labels.

APPROACH

We investigated self-supervised learning (SSL), a promising technique for discovering structure in unlabeled data, to learn representations of EEG signals. Specifically, we explored two tasks based on temporal context prediction as well as contrastive predictive coding on two clinically-relevant problems: EEG-based sleep staging and pathology detection. We conducted experiments on two large public datasets with thousands of recordings and performed baseline comparisons with purely supervised and hand-engineered approaches.

MAIN RESULTS

Linear classifiers trained on SSL-learned features consistently outperformed purely supervised deep neural networks in low-labeled data regimes while reaching competitive performance when all labels were available. Additionally, the embeddings learned with each method revealed clear latent structures related to physiological and clinical phenomena, such as age effects.

SIGNIFICANCE

We demonstrate the benefit of SSL approaches on EEG data. Our results suggest that self-supervision may pave the way to a wider use of deep learning models on EEG data.

Decoding emotional valence from electroencephalographic rhythmic activity

We attempt to decode emotional valence from electroencephalographic rhythmic activity in a naturalistic setting. We employ a data-driven method developed in a previous study, Spectral Linear Discriminant Analysis, to discover the relationships between the classification task and independent neuronal sources, optimally utilizing multiple frequency bands. A detailed investigation of the classifier provides insight into the neuronal sources related with emotional valence, and the individual differences of the subjects in processing emotions. Our findings show: (1) sources whose locations are similar across subjects are consistently involved in emotional responses, with the involvement of parietal sources being especially significant, and (2) even though the locations of the involved neuronal sources are consistent, subjects can display highly varying degrees of valence-related EEG activity in the sources.

Independent component analysis of fMRI group studies by self-organizing clustering

Independent component analysis (ICA) is a valuable technique for the multivariate data-driven analysis of functional magnetic resonance imaging (fMRI) data sets. Applications of ICA have been developed mainly for single subject studies, although different solutions for group studies have been proposed. These approaches combine data sets from multiple subjects into a single aggregate data set before ICA estimation and, thus, require some additional assumptions about the separability across subjects of group independent components. Here, we exploit the application of similarity measures and a related visual tool to study the natural self-organizing clustering of many independent components from multiple individual data sets in the subject space. Our proposed framework flexibly enables multiple criteria for the generation of group independent components and their random-effects evaluation. We present real visual activation fMRI data from two experiments, with different spatiotemporal structures, and demonstrate the validity of this framework for a blind extraction and selection of meaningful activity and functional connectivity group patterns. Our approach is either alternative or complementary to the group ICA of aggregated data sets in that it exploits commonalities across multiple subject-specific patterns, while addressing as much as possible of the intersubject variability of the measured responses. This property is particularly of interest for a blind group and subgroup pattern extraction and selection.

Sparse code shrinkage: denoising of nongaussian data by maximum likelihood estimation

Sparse coding is a method for finding a representation of data in which each of the components of the representation is only rarely significantly active. Such a representation is closely related to redundancy reduction and independent component analysis, and has some neurophysiological plausibility. In this article, we show how sparse coding can be used for denoising. Using maximum likelihood estimation of nongaussian variables corrupted by gaussian noise, we show how to apply a soft-thresholding (shrinkage) operator on the components of sparse coding so as to reduce noise. Our method is closely related to the method of wavelet shrinkage, but it has the important benefit over wavelet methods that the representation is determined solely by the statistical properties of the data. The wavelet representation, on the other hand, relies heavily on certain mathematical properties (like self-similarity) that may be only weakly related to the properties of natural data.