Objective . Brain–computer interface (BCI) refers to procedures that link the central nervous system to a device. BCI was historically performed using electroencephalography (EEG). In the last years, encouraging results were obtained by combining EEG with other neuroimaging technologies, such as functional near infrared spectroscopy (fNIRS). A crucial step of BCI is brain state classification from recorded signal features. Deep artificial neural networks (DNNs) recently reached unprecedented complex classification outcomes. These performances were achieved through increased computational power, efficient learning algorithms, valuable activation functions, and restricted or back-fed neurons connections. By expecting significant overall BCI performances, we investigated the capabilities of combining EEG and fNIRS recordings with state-of-the-art deep learning procedures. Approach . We performed a guided left and right hand motor imagery task on 15 subjects with a fixed classification response time of 1 s and overall experiment length of 10 min. Left versus right classification accuracy of a DNN in the multi-modal recording modality was estimated and it was compared to standalone EEG and fNIRS and other classifiers. Main results . At a group level we obtained significant increase in performance when considering multi-modal recordings and DNN classifier with synergistic effect. Significance . BCI performances can be significantly improved by employing multi-modal recordings that provide electrical and hemodynamic brain activity information, in combination with advanced non-linear deep learning classification procedures.

Over the last decade, EEG resting-state microstate analysis has evolved from a niche existence to a widely used and well-accepted methodology. The rapidly increasing body of empirical findings started to yield overarching patterns of associations of biological and psychological states and traits with specific microstate classes. However, currently, this cross-referencing among apparently similar microstate classes of different studies is typically done by “eyeballing” of printed template maps by the individual authors, lacking a systematic procedure. To improve the reliability and validity of future findings, we present a tool to systematically collect the actual data of template maps from as many published studies as possible and present them in their entirety as a matrix of spatial similarity. The tool also allows importing novel template maps and systematically extracting the findings associated with specific microstate maps from ongoing or published studies. The tool also allows importing novel template maps and systematically extracting the findings associated with specific microstate maps in the literature. The analysis of 40 included sets of template maps indicated that: (i) there is a high degree of similarity of template maps across studies, (ii) similar template maps were associated with converging empirical findings, and (iii) representative meta-microstates can be extracted from the individual studies. We hope that this tool will be useful in coming to a more comprehensive, objective, and overarching representation of microstate findings.

Neonatal seizures are the most common clinical presentation of neurological conditions and can have adverse effects on the neurodevelopment of the neonatal brain. Visual detection of these events from continuous EEG recordings is a laborious and time-consuming task. We propose a novel algorithm for the automated detection of neonatal seizures. In this study, we propose a novel deep learning model based on Graph Convolutional Neural Networks for the automated detection of neonatal seizures. Unlike other methods exploiting mainly the temporal information contained in EEG signals, our method also considers long-range spatial information, i.e., the interdependencies across EEG signals. The temporal information is embedded as graph signals in the graph representation of the EEG recordings and includes EEG features extracted from the EEG signals in the time and frequency domains. The spatial information is represented as functional connections among the EEG channels (calculated by the phase-locking value and the mean squared coherence) or as maps of Euclidean distances. These different spatial representations were evaluated to assess their efficiency in providing more discriminative features for an effective detection of neonatal seizures. The model performance was assessed on a publicly available dataset of continuous EEG signals recorded from 39 neonates by means of the area under the curve (AUC) and the AUC for specificity values greater than 90% (AUC90). After applying post-processing, consisting in smoothing the output of the classifiers, the models based on the mean squared coherence, the phase-locking value, and the Euclidean distance respectively reached a median AUC of 99.1% (IQR: 96.8%-99.6%), 99% (IQR: 95.2%-99.7%), and 97.3% (IQR: 86.3%-99.6%), and a median AUC90 of 96%, 95.7%, and 94.9%. These values are superior or comparable to those reached by methods considered as state-of-the-art in this field. Our results show that the EEG graph representations drawn from functional connectivity measures can effectively leverage interdependencies among EEG signals and lead to reliable detection of neonatal seizures. Furthermore, our model has the advantage of requiring only temporal annotations on seizures for the training phase, making it more appealing for clinical applications.

The interference effects of transcranial magnetic stimulation (TMS) on several electroencephalographic (EEG) measures in both temporal and frequency domains have been reported. We tested the hypothesis whether the offline external inhibitory interference, although focal, could result in a global reorganization of the functional brain state, as assessed by EEG microstates. In 16 healthy subjects, we inhibited five parietal areas and used a pseudo stimulation (Sham) at rest. The EEG microstates were extracted before and after each stimulation. The canonical A, B, C and D templates were found before and after all stimulation conditions. The Sham, as well as the stimulation of a ventral site did not modify any resting EEG microstates’ topography. On the contrary, interfering with parietal key-nodes of both dorsal attention (DAN) and default mode networks (DMN), we observed that the microstate C clearly changes, whereas the other three topographies are not affected. These results provide the first causal evidence of a microstates modification following magnetic interference. Since the microstate C has been associated to the activity in regions belonging to the cingulo-opercular network (CON), the regional specificity of such inhibition seems to support the theory of a link between CON and both DAN and DMN at rest.

Objective: Interpretation of the Electroencephalographic (EEG) and Magnetoencephalographic (MEG) signals requires off-line artifacts removal. Since artifacts share frequencies with brain activity, filtering is insufficient. Blind source separation, mainly through Independent Component Analysis (ICA), is the gold-standard procedure for the identification of artifacts in multi-dimensional recordings. However, a classification of brain and artifactual Independent Components (ICs) is still required. Since ICs exhibit recognizable patterns, classification is usually performed by experts' visual inspection. This procedure is time-consuming and prone to errors. Automatic ICs classification has been explored, often through complex ICs features extraction prior to classification. Relying on Deep Learning ability of self-extracting the features of interest, we investigated the capabilities of Convolutional Neural Networks (CNNs) for off-line, automatic artifact identification through ICs without feature selection. Methods: A CNN was applied to spectrum and topography of a large data-set of few thousand samples of ICs obtained from multi-channel EEG and MEG recordings acquired during heterogeneous experimental settings and on different subjects. CNN performances, when applied to EEG, MEG, and combined EEG and MEG ICs, were explored and compared with state of the art feature-based automatic classification. Results: Beyond state-of-the-art automatic classification accuracies were demonstrated through cross-validation (92.4% EEG, 95.4% MEG, 95.6% EEG+MEG). Conclusion: High CNN classification performances were achieved through heuristically selection of machinery hyperparameters and through the CNN self-selection of the features of interest. Significance: Considering the large data availability of multi-channel EEG and MEG recordings, CNNs may be suited for classification of ICs of multi-channel brain electrophysiological recordings.