EEGminer: Discovering Interpretable Features of Brain Activity with Learnable Filters

Abstract

Patterns of brain activity are associated with different brain processes and can be used to identify different brain states and make behavioral predictions. However, the relevant features are not readily apparent and accessible. To mine informative latent representations from multichannel EEG recordings, we propose a novel differentiable EEG decoding pipeline consisting of learnable filters and a pre-determined feature extraction module. Specifically, we introduce filters parameterized by generalized Gaussian functions that offer a smooth derivative for stable end-to-end model training and allow for learning interpretable features. For the feature module, we use signal magnitude and functional connectivity. We demonstrate the utility of our model towards emotion recognition from EEG signals on the SEED dataset, as well as on a new EEG dataset of unprecedented size (i.e., 763 subjects), where we identify consistent trends of music perception and related individual differences. The discovered features align with previous neuroscience studies and offer new insights, such as marked differences in the functional connectivity profile between left and right temporal areas during music listening. This agrees with the respective specialisation of the temporal lobes regarding music perception proposed in the literature.

BrainWave-Scattering Net: a lightweight network for EEG-based motor imagery recognition

“…Finally, in the area of EEG signal analysis, learnable filters have also been used to enhance the interpretability properties of DL architectures like [44] where a SincNet-based classifier is used for emotional EEG signals and [45] where trainable generalized Gaussian filters are used for discovering explicit features of spontaneous brain activity.…”

“…Before extracting trials from these continuous mode recordings, a series of 'data-cleaning' steps was performed so as to reduce the effects of contamination by biological artifacts. This step was dictated by the fact that only a small number of trials (45) was available for each participant, and we had to assure sufficient information for training (and validating) personalized models. The preprocessing included the following steps:…”

An AI-Inspired Spatio-Temporal Neural Network for EEG-Based Emotional Status

“…The discrete wavelet transform was employed in [ 34 ] to identify and filter out unwanted frequency components from the raw EEGs. Learnable temporal filters were introduced in [ 35 ] that learn and optimize the filter weights and parameters for EEG-based brain activity pattern identification. Instead of raw EEGs, the spectrogram of an EEG using Short-Time Fourier Transform (STFT) [ 36 ] was employed as a pre-processing method to enhance the classification accuracy of the model.…”