EEG and Brain-Computer Interfaces

This study evaluates brain-wide regions for speech neuroprostheses, finding that sulcal depth areas provide better decoding accuracy than the motor cortex.

This paper develops an AI-enhanced non-invasive brain-machine interface (BMI) that improves control performance using a CNN and Kalman filter, enabling users to control cursors and robotic arms.

This study analyzes neural activity in the precentral gyrus of two spinal cord injury patients to classify hand gestures with 70% accuracy, showing potential for intracortical BCI applications.

This paper applies hypergraph models to analyze multidimensional neuroscience data, revealing distinct connectivity patterns in sleep stages and epilepsy types.

This paper evaluates deep learning methods for sleep stage classification from EEG signals, finding that LSTM-based models provide more consistent trustworthiness and calibration than EEGNet.

The paper demonstrates that 40 Hz SSVEPs can be reliably measured during frequency-matched electrical brain stimulation, distinguishing neuronal activity from electrical artefacts.

The paper evaluates correlation-based features for neural voice activity detection using ECoG signals from an ALS patient, finding they outperform traditional log-power features in long-term stability.

The paper introduces a source-level noise covariance model for resting-state MEG/EEG data, improving source reconstruction reliability by avoiding artificial baseline distortions.

This paper proposes using Schrödinger bridge to generate images from EEG by reversing noise accumulation, improving robustness to inter-subject variability compared to diffusion models.

The paper semi-automatically annotates ictal chirps in iEEG data from epilepsy patients and finds their spectro-temporal features correlate with seizure onset zones and surgical outcomes.

This paper decodes intended speech from a person with anarthria and locked-in syndrome using an intracortical brain-computer interface, showing that phonemes, words, and higher-order language units can be decoded above chance.

KenazLBM is a model that characterizes and predicts brain states from raw iEEG data, generalizing across subjects with heterogeneous recording channels.

This paper applies Takens' theorem to EEG data to reveal regional differences in brain dynamics across frontal, sensorimotor, and occipital regions.

Concept2Brain generates synthetic electrophysiological responses to text and pictures using a deep learning model, enabling reproducible EEG datasets and AI-driven brain activity modeling.

This study compares traditional and BOSC methods for sleep spindle detection, finding BOSC more accurate in identifying valid spindles and reducing false detections.

This study uses EEG data and a large language model to reconstruct imagined speech and estimate metabolic activity, showing improved biological variability with pseudorandom variance.

This study proposes optimized dry EEG electrode tip geometries to reduce mechanical stress on the scalp under tilted conditions using finite element analysis.

DynamoSort automatically classifies seizure dynamotypes using machine learning, achieving performance comparable to human raters and providing probabilistic scores for seizure dynamics.

This paper introduces two wireless behind-the-ear EEG sensors that synchronize data via a USB dongle, enabling improved signal processing for auditory steady-state response detection.

This study quantifies how defacing MRI data affects EEG source localization accuracy using boundary and finite element modeling in ten subjects.

The paper investigates absolute auditory attention decoding using EEG, finding that the 0.1-1 Hz band better distinguishes attentional states than the 1-4 Hz band.

This study shows that beta power reflects motor flexibility rather than vigor, as reducing beta power improved task performance across different motor tasks.

This paper identifies six critical limitations of Kalman filters in motor BMI by violating neuroscience principles and proposes a perspective shift to improve performance.

This paper introduces a functionally-structured Bayesian model to improve MEG source estimation accuracy and reduce information spreading.

The paper demonstrates the feasibility of collecting high-quality EEG data in field settings in India and Tanzania using trained teams and automated quality control, achieving cost-effective large-scale data collection.

EEG-GAN generates synthetic EEG data to improve classification accuracy across multiple datasets and classifiers, achieving up to 16% improvement in decoding accuracy.

This study uses a biophysical model to investigate how neural mechanisms change during BCI training, finding training-induced excitatory-inhibitory reconfiguration in sensorimotor areas.

This study uses EEG and a humanoid robot to predict human and robot movement during motor learning, showing neural correlates of learning and decoding movement parameters.

The paper presents two EEG datasets for speech decoding, focusing on phoneme and word classification and articulatory properties, with conditions involving transcranial magnetic stimulation.