A causal perspective on brainwave modeling for brain-computer interfaces.

Objective. Machine learning (ML) models have opened up enormous opportunities in the field of brain-computer Interfaces (BCIs). Despite their great success, they usually face severe limitations when they are employed in real-life applications outside a controlled laboratory setting.Approach. Mixing causal reasoning, identifying causal relationships between variables of interest, with brainwave modeling can change one's viewpoint on some of these major challenges which can be found in various stages in the ML pipeline, ranging from data collection and data pre-processing to training methods and techniques.Main results. In this work, we employ causal reasoning and present a framework aiming to breakdown and analyze important challenges of brainwave modeling for BCIs.Significance. Furthermore, we present how general ML practices as well as brainwave-specific techniques can be utilized and solve some of these identified challenges. And finally, we discuss appropriate evaluation schemes in order to measure these techniques' performance and efficiently compare them with other methods that will be developed in the future.

EEGminer: discovering interpretable features of brain activity with learnable filters.

Objective. The patterns of brain activity associated with different brain processes can be used to identify different brain states and make behavioural predictions. However, the relevant features are not readily apparent and accessible. Our aim is to design a system for learning informative latent representations from multichannel recordings of ongoing EEG activity.Approach: We propose a novel differentiable 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 estimates.Main results.We demonstrate the utility of our model on a new EEG dataset of unprecedented size (i.e. 721 subjects), where we identify consistent trends of music perception and related individual differences. Furthermore, we train and apply our model in two additional datasets, specifically for emotion recognition on SEED and workload classification on simultaneous task EEG workload. The discovered features align well 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 specialisation of the temporal lobes regarding music perception proposed in the literature.Significance. The proposed method offers strong interpretability of learned features while reaching similar levels of accuracy achieved by black box deep learning models. This improved trustworthiness may promote the use of deep learning models in real world applications. The model code is available athttps://github.com/SMLudwig/EEGminer/.

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

Objective.Brain-computer interfaces (BCIs) enable a direct communication of the brain with the external world, using one's neural activity, measured by electroencephalography (EEG) signals. In recent years, convolutional neural networks (CNNs) have been widely used to perform automatic feature extraction and classification in various EEG-based tasks. However, their undeniable benefits are counterbalanced by the lack of interpretability properties as well as the inability to perform sufficiently when only limited amount of training data is available.Approach.In this work, we introduce a novel, lightweight, fully-learnable neural network architecture that relies on Gabor filters to delocalize EEG signal information into scattering decomposition paths along frequency and slow-varying temporal modulations.Main results.We utilize our network in two distinct modeling settings, for building either a generic (training across subjects) or a personalized (training within a subject) classifier.Significance.In both cases, using two different publicly available datasets and one in-house collected dataset, we demonstrate high performance for our model with considerably less number of trainable parameters as well as shorter training time compared to other state-of-the-art deep architectures. Moreover, our network demonstrates enhanced interpretability properties emerging at the level of the temporal filtering operation and enables us to train efficient personalized BCI models with limited amount of training data.

Improving Generalization of CNN-Based Motor-Imagery EEG Decoders via Dynamic Convolutions.

Deep Convolutional Neural Networks (CNNs) have recently demonstrated impressive results in electroencephalogram (EEG) decoding for several Brain-Computer Interface (BCI) paradigms, including Motor-Imagery (MI). However, neurophysiological processes underpinning EEG signals vary across subjects causing covariate shifts in data distributions and hence hindering the generalization of deep models across subjects. In this paper, we aim to address the challenge of inter-subject variability in MI. To this end, we employ causal reasoning to characterize all possible distribution shifts in the MI task and propose a dynamic convolution framework to account for shifts caused by the inter-subject variability. Using publicly available MI datasets, we demonstrate improved generalization performance (up to 5%) across subjects in various MI tasks for four well-established deep architectures.

On the estimate of music appraisal from surface EEG: a dynamic-network approach based on cross-sensor PAC measurements.

Objective.The aesthetic evaluation of music is strongly dependent on the listener and reflects manifold brain processes that go well beyond the perception of incident sound. Being a high-level cognitive reaction, it is difficult to predict merely from the acoustic features of the audio signal and this poses serious challenges to contemporary music recommendation systems. We attempted to decode music appraisal from brain activity, recorded via wearable EEG, during music listening.Approach.To comply with the dynamic nature of music stimuli, cross-frequency coupling measurements were employed in a time-evolving manner to capture the evolving interactions between distinct brain-rhythms during music listening. Brain response to music was first represented as a continuous flow of functional couplings referring to both regional and inter-regional brain dynamics and then modelled as an ensemble of time-varying (sub)networks. Dynamic graph centrality measures were derived, next, as the final feature-engineering step and, lastly, a support-vector machine was trained to decode the subjective music appraisal. A carefully designed experimental paradigm provided the labeled brain signals.Main results.Using data from 20 subjects, dynamic programming to tailor the decoder to each subject individually and cross-validation, we demonstrated highly satisfactory performance (MAE= 0.948,R2= 0.63) that can be attributed, mostly, to interactions of left frontal gamma rhythm. In addition, our music-appraisal decoder was also employed in a part of the DEAP dataset with similar success. Finally, even a generic version of the decoder (common for all subjects) was found to perform sufficiently.Significance.A novel brain signal decoding scheme was introduced and validated empirically on suitable experimental data. It requires simple operations and leaves room for real-time implementation. Both the code and the experimental data are publicly available.

Covariation Informed Graph Slepians for Motor Imagery Decoding.

Graph signal processing (GSP) provides signal analytic tools for data defined in irregular domains, as is the case of non-invasive electroencephalography (EEG). In this work, the recently introduced technique of Graph Slepian functions is exploited for the robust decoding of motor imagery (MI) brain activity. The particular technique builds over the concept of graph Fourier transform (GFT) and provides additional flexibility in the subsequent data analysis by incorporating domain knowledge. Based on contrastive learning, we introduce an algorithmic pipeline that attains a data driven and subject specific design of Graph Slepian functions. These functions, by incorporating both the topology of the sensor array and the empirical evidence about the differential functional covariation, act as spatial filters that enhance the information conveyed by the multichannel signal and specifically relates to the participant's intention. The proposed technique for crafting Graph Slepians is incorporated in a MI-decoding scheme, in which the informed projections are fed to a support vector machine (SVM) that casts a prediction regarding the type of intended movement. The employed MI-decoder is evaluated based on two publicly available datasets and its superiority against popular alternatives in the field is established. Computational efficiency is listed among its main advantages, since it involves only simple matrix operations, allowing to consider its use in real-time implementations.

A complex-valued functional brain connectivity descriptor amenable to Riemannian geometry.

OBJECTIVE: We introduce a novel, phase-based, functional connectivity descriptor that encapsulates not only the synchronization strength between distinct brain regions, but also the time-lag between the involved neural oscillations. The new estimator employs complex-valued measurements and results in a brain network sketch that lives on the smooth manifold of Hermitian Positive Definite (HPD) matrices. APPROACH: Leveraging the HPD property of the proposed descriptor, we adapt a recently introduced dimensionality reduction methodology that is based on Riemannian Geometry and discriminatively detects the recording sites which best reflect the differences in network organization between contrasting recording conditions in order to overcome the problem of high-dimensionality, usually encountered in the connectivity patterns derived from multisite encephalographic recordings. MAIN RESULTS: The proposed framework is validated using an EEG dataset that refers to the challenging problem of differentiating between attentive and passive visual responses. We provide evidence that the reduced connectivity representation facilitates high classification performance and caters for neuroscientific explorations. SIGNIFICANCE: Our paper is the very first that introduces an advanced connectivity descriptor that can take advantage of Riemannian geometry tools. The proposed descriptor, that inherently and simultaneously captures both the strength and the corresponding time-lag of the phase synchronization, is the first phase-based descriptor tailored to leverage the benefits of Riemannian geometry.

Using Discriminative Lasso to Detect a Graph Fourier Transform (GFT) Subspace for robust decoding in Motor Imagery BCI.

A novel decoding scheme for motor imagery (MI) brain computer interfaces (BCI's) is introduced based on the GFT concept. It considers the recorded EEG activity as a signal defined over (the graph of) the sensor array. A graph encapsulating the functional covariations emerging during the execution of a specific imagined movement is first defined, from a small training set of relevant trials. The ensemble of graphs signals corresponding to a multi-trial training dataset is then analyzed using a graph-guided decomposition and, based on discriminative Lasso (dLasso), an information-rich GFT subspace is defined. After training, only simple matrix operations are required for transforming the multichannel signal into features to be fed into a classifier that decides whether brain activity conforms with the graph structure associated with the targeted movement. The proposed decoding scheme is evaluated based on two different datasets and found to compare favorably against popular alternatives in the field.

Harnessing functional segregation across brain rhythms as a means to detect EEG oscillatory multiplexing during music listening.

OBJECTIVE: Music, being a multifaceted stimulus evolving at multiple timescales, modulates brain function in a manifold way that encompasses not only the distinct stages of auditory perception, but also higher cognitive processes like memory and appraisal. Network theory is apparently a promising approach to describe the functional reorganization of brain oscillatory dynamics during music listening. However, the music induced changes have so far been examined within the functional boundaries of isolated brain rhythms. APPROACH: Using naturalistic music, we detected the functional segregation patterns associated with different cortical rhythms, as these were reflected in the surface electroencephalography (EEG) measurements. The emerged structure was compared across frequency bands to quantify the interplay among rhythms. It was also contrasted against the structure from the rest and noise listening conditions to reveal the specific components stemming from music listening. Our methodology includes an efficient graph-partitioning algorithm, which is further utilized for mining prototypical modular patterns, and a novel algorithmic procedure for identifying 'switching nodes' (i.e. recording sites) that consistently change module during music listening. MAIN RESULTS: Our results suggest the multiplex character of the music-induced functional reorganization and particularly indicate the dependence between the networks reconstructed from the δ and ß H rhythms. This dependence is further justified within the framework of nested neural oscillations and fits perfectly within the context of recently introduced cortical entrainment to music. SIGNIFICANCE: Complying with the contemporary trends towards a multi-scale examination of the brain network organization, our approach specifies the form of neural coordination among rhythms during music listening. Considering its computational efficiency, and in conjunction with the flexibility of in situ electroencephalography, it may lead to novel assistive tools for real-life applications.

Spontaneous Up states in vitro: a single-metric index of the functional maturation and regional differentiation of the cerebral cortex.

Understanding the development and differentiation of the neocortex remains a central focus of neuroscience. While previous studies have examined isolated aspects of cellular and synaptic organization, an integrated functional index of the cortical microcircuit is still lacking. Here we aimed to provide such an index, in the form of spontaneously recurring periods of persistent network activity -or Up states- recorded in mouse cortical slices. These coordinated network dynamics emerge through the orchestrated regulation of multiple cellular and synaptic elements and represent the default activity of the cortical microcircuit. To explore whether spontaneous Up states can capture developmental changes in intracortical networks we obtained local field potential recordings throughout the mouse lifespan. Two independent and complementary methodologies revealed that Up state activity is systematically modified by age, with the largest changes occurring during early development and adolescence. To explore possible regional heterogeneities we also compared the development of Up states in two distinct cortical areas and show that primary somatosensory cortex develops at a faster pace than primary motor cortex. Our findings suggest that in vitro Up states can serve as a functional index of cortical development and differentiation and can provide a baseline for comparing experimental and/or genetic mouse models.

In quest of the missing neuron: spike sorting based on dominant-sets clustering.

Spike sorting algorithms aim at decomposing complex extracellular signals to independent events from single neurons in the electrode's vicinity. The decision about the actual number of active neurons is still an open issue, with sparsely firing neurons and background activity the most influencing factors. We introduce a graph-theoretical algorithmic procedure that successfully resolves this issue. Dimensionality reduction coupled with a modern, efficient and progressively executable clustering routine proved to achieve higher performance standards than popular spike sorting methods. Our method is validated extensively using simulated data for different levels of SNR.

NASS: an empirical approach to spike sorting with overlap resolution based on a hybrid noise-assisted methodology.

Background noise and spike overlap pose problems in contemporary spike-sorting strategies. We attempted to resolve both issues by introducing a hybrid scheme that combines the robust representation of spike waveforms to facilitate the reliable identification of contributing neurons with efficient data learning to enable the precise decomposition of coactivations. The isometric feature mapping (ISOMAP) technique reveals the intrinsic data structure, helps with recognising the involved neurons and, simultaneously, identifies the overlaps. Exemplar activation patterns are first estimated for all detected neurons and consecutively used to build a synthetic database in which spike overlaps are systematically varied and realistic noise is added. An Extreme Learning Machine (ELM) is then trained with the ISOMAP representation of this database and learns to associate the synthesised waveforms with the corresponding source neurons. The trained ELM is finally applied to the actual overlaps from the experimental data and this completes the entire spike-sorting process. Our approach is better characterised as semi-supervised, noise-assisted strategy of an empirical nature. The user's engagement is restricted at recognising the number of active neurons from low-dimensional point-diagrams and at deciding about the complexity of overlaps. Efficiency is inherited from the incorporation of well-established algorithms. Moreover, robustness is guaranteed by adaptation to the actual noise properties of a given data set. The validity of our work has been verified via extensive experimentation, using realistically simulated data, under different levels of noise.

Performance evaluation of PCA-based spike sorting algorithms.

Deciphering the electrical activity of individual neurons from multi-unit noisy recordings is critical for understanding complex neural systems. A widely used spike sorting algorithm is being evaluated for single-electrode nerve trunk recordings. The algorithm is based on principal component analysis (PCA) for spike feature extraction. In the neuroscience literature it is generally assumed that the use of the first two or most commonly three principal components is sufficient. We estimate the optimum PCA-based feature space by evaluating the algorithm's performance on simulated series of action potentials. A number of modifications are made to the open source nev2lkit software to enable systematic investigation of the parameter space. We introduce a new metric to define clustering error considering over-clustering more favorable than under-clustering as proposed by experimentalists for our data. Both the program patch and the metric are available online. Correlated and white Gaussian noise processes are superimposed to account for biological and artificial jitter in the recordings. We report that the employment of more than three principal components is in general beneficial for all noise cases considered. Finally, we apply our results to experimental data and verify that the sorting process with four principal components is in agreement with a panel of electrophysiology experts.