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

Journal of Neural Engineering, Journal Year: 2020, Volume and Issue: 18(4), P. 046020 - 046020

Published: Nov. 12, 2020

Objective.Supervised learning paradigms are often limited by the amount of labeled data that is available. This phenomenon particularly problematic in clinically-relevant data, such as electroencephalography (EEG), where labeling can be costly terms specialized expertise and human processing time. Consequently, deep architectures designed to learn on EEG have yielded relatively shallow models performances at best similar those traditional feature-based approaches. However, most situations, unlabeled available abundance. By extracting information from this it might possible reach competitive performance with neural networks despite access labels.Approach.We investigated self-supervised (SSL), a promising technique for discovering structure representations signals. Specifically, we explored two tasks based temporal context prediction well contrastive predictive coding problems: EEG-based sleep staging pathology detection. We conducted experiments large public datasets thousands recordings performed baseline comparisons purely supervised hand-engineered approaches.Main results.Linear classifiers trained SSL-learned features consistently outperformed low-labeled regimes while reaching when all labels were Additionally, embeddings learned each method revealed clear latent structures related physiological clinical phenomena, age effects.Significance.We demonstrate benefit SSL approaches data. Our results suggest self-supervision may pave way wider use

Language: Английский

---

BENDR: Using Transformers and a Contrastive Self-Supervised Learning Task to Learn From Massive Amounts of EEG Data

Frontiers in Human Neuroscience, Journal Year: 2021, Volume and Issue: 15

Published: June 23, 2021

Deep neural networks (DNNs) used for brain–computer interface (BCI) classification are commonly expected to learn general features when trained across a variety of contexts, such that these could be fine-tuned specific contexts. While some success is found in an approach, we suggest this interpretation limited and alternative would better leverage the newly (publicly) available massive electroencephalography (EEG) datasets. We consider how adapt techniques architectures language modeling (LM) appear capable ingesting awesome amounts data toward development encephalography with DNNs same vein. specifically approach effectively automatic speech recognition, which similarly (to LMs) uses self-supervised training objective compressed representations raw signals. After adaptation EEG, find single pre-trained model completely novel EEG sequences recorded differing hardware, different subjects performing tasks. Furthermore, both internal entire architecture can downstream BCI tasks, outperforming prior work more task-specific (sleep stage classification) self-supervision.

Language: Английский

---

Exploring Convolutional Neural Network Architectures for EEG Feature Extraction

Sensors, Journal Year: 2024, Volume and Issue: 24(3), P. 877 - 877

Published: Jan. 29, 2024

The main purpose of this paper is to provide information on how create a convolutional neural network (CNN) for extracting features from EEG signals. Our task was understand the primary aspects creating and fine-tuning CNNs various application scenarios. We considered characteristics signals, coupled with an exploration signal processing data preparation techniques. These techniques include noise reduction, filtering, encoding, decoding, dimension among others. In addition, we conduct in-depth analysis well-known CNN architectures, categorizing them into four distinct groups: standard implementation, recurrent convolutional, decoder architecture, combined architecture. This further offers comprehensive evaluation these covering accuracy metrics, hyperparameters, appendix that contains table outlining parameters commonly used architectures feature extraction

Language: Английский

---

GREEN: A lightweight architecture using learnable wavelets and Riemannian geometry for biomarker exploration with EEG signals

Patterns, Journal Year: 2025, Volume and Issue: 6(3), P. 101182 - 101182

Published: Feb. 13, 2025

Language: Английский

---

Combining magnetoencephalography with magnetic resonance imaging enhances learning of surrogate-biomarkers

eLife, Journal Year: 2020, Volume and Issue: 9

Published: May 19, 2020

Electrophysiological methods, that is M/EEG, provide unique views into brain health. Yet, when building predictive models from data, it often unclear how electrophysiology should be combined with other neuroimaging methods. Information can redundant, useful common representations of multimodal data may not obvious and collection medically contraindicated, which reduces applicability. Here, we propose a model to robustly combine MEG, MRI fMRI for prediction. We focus on age prediction as surrogate biomarker in 674 subjects the Cam-CAN dataset. Strikingly, showed additive effects supporting distinct brain-behavior associations. Moreover, contribution MEG was best explained by cortical power spectra between 8 30 Hz. Finally, demonstrate preserves benefits stacking some missing. The proposed framework, hence, enables learning wide range biomarkers diverse types signals.

Language: Английский

---

Automatic detection of abnormal EEG signals using wavelet feature extraction and gradient boosting decision tree

Biomedical Signal Processing and Control, Journal Year: 2021, Volume and Issue: 70, P. 102957 - 102957

Published: July 15, 2021

Language: Английский

---

EEG-ITNet: An Explainable Inception Temporal Convolutional Network for Motor Imagery Classification

IEEE Access, Journal Year: 2022, Volume and Issue: 10, P. 36672 - 36685

Published: Jan. 1, 2022

In recent years, neural networks and especially deep architectures have received substantial attention for EEG signal analysis in the field of brain-computer interfaces (BCIs). this ongoing research area, end-to-end models are more favoured than traditional approaches requiring transformation pre-classification. They can eliminate need prior information from experts extraction handcrafted features. However, although several learning algorithms been already proposed literature, achieving high accuracies classifying motor movements or mental tasks, they often face a lack interpretability therefore not quite by neuroscience community. The reasons behind issue be number parameters sensitivity to capture tiny yet unrelated discriminative We propose an architecture called EEG-ITNet comprehensible method visualise network learned patterns. Using inception modules causal convolutions with dilation, our model extract rich spectral, spatial, temporal multi-channel signals less complexity (in terms trainable parameters) other existing architectures, such as EEG-Inception EEG-TCNet. By exhaustive evaluation on dataset 2a BCI competition IV OpenBMI imagery dataset, shows up 5.9\% improvement classification accuracy different scenarios statistical significance compared its competitors. also comprehensively explain support validity illustration neuroscientific perspective. made code open at https://github.com/AbbasSalami/EEG-ITNet

Language: Английский

---

A reusable benchmark of brain-age prediction from M/EEG resting-state signals

NeuroImage, Journal Year: 2022, Volume and Issue: 262, P. 119521 - 119521

Published: July 26, 2022

Population-level modeling can define quantitative measures of individual aging by applying machine learning to large volumes brain images. These age, obtained from the general population, helped characterize disease severity in neurological populations, improving estimates diagnosis or prognosis. Magnetoencephalography (MEG) and Electroencephalography (EEG) have potential further generalize this approach towards prevention public health enabling assessments at scales socioeconomically diverse environments. However, more research is needed methods that handle complexity diversity M/EEG signals across real-world contexts. To catalyse effort, here we propose reusable benchmarks competing approaches for age modeling. We benchmarked popular classical pipelines deep architectures previously used pathology decoding estimation 4 international cohorts countries cultural contexts, including recordings than 2500 participants. Our were built on top adaptations BIDS standard, providing tools be applied with minimal modification any dataset provided format. results suggest that, regardless whether was used, highest performance reached involving spatially aware representations signals, leading R2 scores between 0.60-0.74. Hand-crafted features paired random forest regression robust even situations which other failed. Taken together, set benchmarks, accompanied open-source software high-level Python scripts, serve as a starting point reference future efforts developing M/EEG-based aging. The generality renders benchmark related objectives such specific cognitive variables clinical endpoints.

Language: Английский

---

The two decades brainclinics research archive for insights in neurophysiology (TDBRAIN) database

Scientific Data, Journal Year: 2022, Volume and Issue: 9(1)

Published: June 14, 2022

In neuroscience, electroencephalography (EEG) data is often used to extract features (biomarkers) identify neurological or psychiatric dysfunction predict treatment response. At the same time neuroscience becoming more data-driven, made possible by computational advances. support of biomarker development and methodologies such as training Artificial Intelligent (AI) networks we present extensive Two Decades-Brainclinics Research Archive for Insights in Neurophysiology (TDBRAIN) EEG database. This clinical lifespan database (5-89 years) contains resting-state, raw EEG-data complemented with relevant demographic a heterogenous collection 1274 patients collected between 2001 2021. Main indications included are Major Depressive Disorder (MDD; N = 426), attention deficit hyperactivity disorder (ADHD; 271), Subjective Memory Complaints (SMC: 119) obsessive-compulsive (OCD; 75). Demographic-, personality- day measurement Thirty percent outcome will remain blinded prospective validation replication purposes. The TDBRAIN code available on Brainclinics Foundation website at www.brainclinics.com/resources Synapse www.synapse.org/TDBRAIN .

Language: Английский

---

Data augmentation for learning predictive models on EEG: a systematic comparison

Journal of Neural Engineering, Journal Year: 2022, Volume and Issue: 19(6), P. 066020 - 066020

Published: Nov. 11, 2022

Objective: The use of deep learning for electroencephalography (EEG) classification tasks has been rapidly growing in the last years, yet its application limited by relatively small size EEG datasets. Data augmentation, which consists artificially increasing dataset during training, can be employed to alleviate this problem. While a few augmentation transformations data have proposed literature, their positive impact on performance is often evaluated single and compared one or two competing methods. This work proposes better validate existing approaches through unified exhaustive analysis. Approach: We compare quantitatively 13 different augmentations with predictive tasks, datasets models, using three types experiments. Main results: demonstrate that employing adequate bring up 45% accuracy improvements low regimes same model trained without any augmentation. Our experiments also show there no best strategy, as good differ each task. Significance: results highlight consider sleep stage motor imagery brain-computer interfaces. More broadly, it demonstrates benefit from

Language: Английский

---