Deep learning is significantly advancing the analysis of electroencephalography (EEG) data by effectively discovering highly nonlinear patterns within the signals. Data partitioning and cross-validation are crucial for assessing model performance and ensuring study comparability, as they can produce varied results and data leakage due to specific signal properties (e.g., biometric). Such variability in model evaluation leads to incomparable studies and, increasingly, overestimated performance claims, which are detrimental to the field. Nevertheless, no comprehensive guidelines for proper data partitioning and cross-validation exist in the domain, nor is there a quantitative evaluation of the impact of different approaches on model accuracy, reliability, and generalizability. To assist researchers in identifying optimal experimental strategies, this paper thoroughly investigates the role of data partitioning and cross-validation in evaluating EEG deep learning models. Five cross-validation settings are compared across three supervised cross-subject classification tasks (brain-computer interfaces, Parkinson's, and Alzheimer's disease classification) and four established architectures of increasing complexity (ShallowConvNet, EEGNet, DeepConvNet, and Temporal-based ResNet). The comparison of over 100,000 trained models underscores, first, the importance of using subject-based cross-validation strategies for evaluating EEG deep learning architectures, except when within-subject analyses are acceptable (e.g., BCI). Second, it highlights the greater reliability of nested approaches (e.g., N-LNSO) compared to non-nested counterparts, which are prone to data leakage and favor larger models overfitting to validation data. In conclusion, this work provides EEG deep learning researchers with an analysis of data partitioning and cross-validation and offers guidelines to avoid data leakage, currently undermining the domain with potentially overestimated performance claims.

Over the last decade, deep learning applications in biomedical research have exploded, demonstrating their ability to often outperform previous machine learning approaches in various tasks. However, training deep learning models for biomedical applications requires large amounts of data annotated by experts, whose collection is often time- and cost- prohibitive. Self-Supervised Learning (SSL) has emerged as a prominent solution for such problem, as it allows to learn powerful representations from vast unlabeled data by producing supervisory signals directly from the data. The high amount of recent works employing the self-supervised learning paradigm for the analysis of biomedical signals (biosignals) can make it difficult for researchers to have a complete picture of the current research state. Therefore, this paper aims at outlining and clarifying the state-of-the-art in the domain. The article: briefly summarizes the nature and acquisition modality of the main biosignals; introduces the self-supervised learning method, focusing on the different pretraining strategies; provides a concise overview of the works employing SSL for the analysis of different types of biosignals; provides an overall analysis of critical aspects to consider when employing SSL to biosignals, also highlighting current open challenges. The analysis of the scientific literature highlights the importance of SSL, confirming its potential to improve models’ performance and robustness, and to promote the integration of deep learning into clinical tasks.

The last decade has witnessed a notable surge in deep learning applications for electroencephalography (EEG) data analysis, showing promising improvements over conventional statistical techniques. However, deep learning models can underperform if trained with bad processed data. Preprocessing is crucial for EEG data analysis, yet there is no consensus on the optimal strategies in deep learning scenarios, leading to uncertainty about the extent of preprocessing required for optimal results. This study is the first to thoroughly investigate the effects of EEG preprocessing in deep learning applications, drafting guidelines for future research. It evaluates the effects of varying preprocessing levels, from raw and minimally filtered data to complex pipelines with automated artifact removal algorithms. Six classification tasks (eye blinking, motor imagery, Parkinson’s, Alzheimer’s disease, sleep deprivation, and first episode psychosis) and four established EEG architectures were considered for the evaluation. The analysis of 4800 trained models revealed statistical differences between preprocessing pipelines at the intra-task level for each model and at the inter-task level for the largest model. Models trained on raw data consistently performed poorly, always ranking last in average scores. In addition, models seem to benefit more from minimal pipelines without artifact handling methods. These findings suggest that EEG artifacts may affect the performance and generalizability of deep neural networks.

SelfEEG is an open-source Python library developed to assist researchers in conducting Self-Supervised Learning (SSL) experiments on electroencephalography (EEG) data. Its primary objective is to offer a user-friendly but highly customizable environment, enabling users to efficiently design and execute self-supervised learning tasks on EEG data.

Objective. This work presents xEEGNet, a novel, compact, and explainable neural network for electroencephalography (EEG) data analysis. It is fully interpretable and reduces overfitting through a major parameter reduction. Approach. As an applicative use case to develop our model, we focused on the classification of common dementia conditions, Alzheimer’s and frontotemporal dementia, versus controls. xEEGNet, however, is broadly applicable to other neurological conditions involving spectral alterations. We used ShallowNet, a simple and popular model in the EEGNet family, as a starting point. Its structure was analyzed and gradually modified to move from a ‘black box’ model to a more transparent one, without compromising performance. The learned kernels and weights were analyzed from a clinical standpoint to assess their medical significance. Model variants, including ShallowNet and the final xEEGNet, were evaluated using a robust nested-leave-n-subjects out cross-validation for unbiased performance estimates. Variability across data splits was explained using the embedded EEG representations, grouped by class and set, with pairwise separability to quantify group distinction. Overfitting was measured through training-validation loss correlation and training speed. Main results. xEEGNet uses only 168 parameters, 200 times fewer than ShallowNet, yet retains interpretability, resists overfitting, achieves comparable median performance (−1.5%), and reduces performance variability across splits. This variability is explained by the embedded EEG representations: higher accuracy correlates with greater separation between test-set controls and Alzheimer’s cases, without significant influence from the training data. Significance. The capability of xEEGNet to filter specific EEG bands, learns band specific topographies and use the right EEG spectral bands for disease classification demonstrates its interpretability. While big deep learning models are typically prioritized for performance, this study shows that smaller architectures like xEEGNet can be equally effective in pathology classification, using EEG data.

Objective. This study aims to address the challenges associated with data-driven electroencephalography (EEG) data analysis by introducing a standardised library called BIDSAlign. This library efficiently processes and merges heterogeneous EEG datasets from different sources into a common standard template. The goal of this work is to create an environment that allows to preprocess public datasets in order to provide data for the effective training of deep learning (DL) architectures. Approach. The library can handle both Brain Imaging Data Structure (BIDS) and non-BIDS datasets, allowing the user to easily preprocess multiple public datasets. It unifies the EEG recordings acquired with different settings by defining a common pipeline and a specified channel template. An array of visualisation functions is provided inside the library, together with a user-friendly graphical user interface to assist non-expert users throughout the workflow. Main results. BIDSAlign enables the effective use of public EEG datasets, providing valuable medical insights, even for non-experts in the field. Results from applying the library to datasets from OpenNeuro demonstrate its ability to extract significant medical knowledge through an end-to-end workflow, facilitating group analysis, visual comparison and statistical testing. Significance. BIDSAlign solves the lack of large EEG datasets by aligning multiple datasets to a standard template. This unlocks the potential of public EEG data for training DL models. It paves the way to promising contributions based on DL to clinical and non-clinical EEG research, offering insights that can inform neurological disease diagnosis and treatment strategies.