Ian Daly

Objective . Semantic brain-computer interfaces (sBCIs) aim to decode conceptual information directly from neural activity, enabling more efficient and natural communication for individuals with severe motor impairments. However, existing approaches are often confounded by stimulus-driven perceptual processes and lack a reliable means of tagging the moment of concept formation. We introduce the graded inventory of semantic triggers (GIST), a novel electroencephalography (EEG) paradigm designed to elicit semantic retrieval from graded verbal clues with precise temporal alignment. Approach . EEG data were recorded from 14 participants performing the GIST task across three semantic categories (food, musical instruments, and body parts). Target neural activity was aligned to each participant’s self-reported moment of semantic retrieval (spacebar press), enabling response-locked analysis. Here, we propose a staged framework where the first comprises a Conformer-based encoder trained on pooled data using a hybrid contrastive objective that combines instance-level consistency and cross-participant class alignment with domain-adversarial learning to reduce participant-specific information. In the second stage, the frozen encoder is transferred to a held-out participant whose data played no part in Stage 1, and a support vector machine is calibrated on that participant’s own labelled items under a nested leave-one-item-out protocol. Results . Above-chance semantic decoding was achieved in all three pairwise tasks. Mean accuracies were 73.4% (Body vs Food), 70.7% (Body vs Music), and 68.2% (Food vs Music). Participant-level permutation tests were significant in 9/11 participants for Body vs Food and 8/11 for each of the other tasks ( p < 0.05). The pre-response interval at [−1.5 s, −0.5 s] was the most discriminative. Significance . The GIST paradigm enables controlled elicitation and temporal alignment of semantic processing while reducing perceptual confounds. Combined with cross-participant representation pretraining and participant-specific calibration, the results demonstrate above-chance decoding of semantic categories from EEG in individuals excluded from representation learning, supporting the feasibility of sBCIs.

Objective. Speech imagery (SI) has emerged as a promising paradigm for brain-computer interface (BCI) control, attracting growing interest due to its intuitive nature–allowing users to interact with the system by internally saying a command. In this study, we investigate the replicability and reproducibility of SI decoding methods. These two aspects are critical in BCI research, where prior literature has highlighted that many studies suffer from incomplete methodological reporting or flawed evaluation procedures, making reproduction difficult. The inherent variability of brain signals further complicates the replication of results. Evaluating the reproducibility of SI decoding approaches is therefore essential to assess the true feasibility of SI as a viable BCI paradigm. Approach. To assess reproducibility, we selected two of the most widely used open-access SI datasets and attempted to reproduce four published decoding pipelines for each dataset. We followed each implementation step-by-step, documented missing or ambiguous information, detailed how we addressed it, and compared our decoding results to those originally reported. To assess replicability, we applied standard decoding pipelines across different time-frequency configurations to three open SI datasets and our own collected dataset. For context and validation, we conducted the same procedure on four publicly available and widely used motor imagery (MI) datasets. Main results. All evaluated SI studies contained some form of missing methodological detail, and others did not include cross-validation procedures. Our reproduction attempts consistently yielded lower classification accuracies than originally reported, with discrepancies ranging from 2 to 39% ( x ~ = 11.25   ± 12.41 % ). In the replication analysis, we found no consistent time-frequency patterns across SI datasets. Furthermore, only 36% of SI participants achieved classification accuracies above statistical significance thresholds, compared to 91% of the participants in MI datasets. Significance. This is the first comprehensive assessment of both reproducibility and replicability in SI decoding. Our findings raise important concerns about the reliability of current SI research and suggest that the feasibility of SI as a practical BCI paradigm may have been overestimated.

Lightweight networks that include depthwise-separable convolution are widely used in motor imagery (MI) electroencephalogram (EEG) decoding of brain–computer interface (BCI). Many established MI classification networks are relatively shallow, preventing them from benefiting from the hierarchical feature extraction capabilities of deeper structures. Due to suboptimal residual connection structures, the mismatched residual baseline layer design, and the poor compatibility between data preprocessing and residual modules, the deepening of networks cannot be effectively combined with residual structures. This creates a depth barrier that hinders further performance improvements. To address these challenges, we propose a novel method, residual depthwise-separable deep neural network (ResDSNet), built upon an unraveled view-path analysis of residual connection structures. The analysis reveals that the residual mechanism achieves optimal performance when the layer distribution across different paths approximates a binomial distribution. Furthermore, we design a residual depthwise-separable convolution module and a tailored data-preprocessing module that effectively integrate with the residual structure, filtering noise and retaining MI task features. We evaluate ResDSNet on three publicly available datasets, including the BCI Competition IV Dataset IIa, the BCI Competition IV Dataset IIb, and the PhysioNet dataset, which collectively contain EEG signals recorded from 127 human subjects. ResDSNet achieves accuracies of 79.36%, 84.95%, and 64.13%, outperforming state-of-the-art methods by 3.16%, 1.59%, and 8.40% with statistical significance. Experimental results indicate that ResDSNet fully unlocks the hierarchical representation capabilities of deep networks for MI-EEG decoding, achieving robust performance and demonstrating substantial potential to overcome the inherent challenges in BCIs.

Steady state visual evoked potential (SSVEP)-based brain-computer interfaces have been widely studied for their fast response speeds and high information transfer rates. However, how to fully utilize the potential information of existing subjects to realize the mining of common information among different subjects and then realize the information migration in a small amount of data scenarios is a difficult problem faced by current research. In order to solve the above problems, this study proposes a deep neural network based on the pyramid squeeze attention (PSA-DNN) mechanism to enhance the performance of SSVEP-BCI through common information migration. Specifically, the band-pass filtered EEG signals were first Fourier transformed to obtain the frequency domain information; subsequently, the frequency domain information is input into a deep neural network, followed by a spatial convolution step to extract spatial domain information. In order to further enhance the quality of information extraction, a pyramid attention module is introduced into the network to realize the enhancement of frequency domain and spatial domain information. Time domain information from the EEG signals is then mined using temporal convolution. Finally, the full connectivity layer is used to output the recognition results. The model is trained in a three-stage stepped approach for SSVEP target recognition. The first stage uses data from all participants in the training set for common information learning and transfers the model parameters trained in the first stage to the network model in the second stage. In the second stage, some of the information from participants in the test set is used for fine-tuning and to mine personalized information from these new participants. The third stage uses the remaining data from participants in the test set to produce classification results. The proposed method is systematically evaluated using the Benchmark and BETA datasets, where it demonstrates favorable performance compared to established baselines. These findings contribute theoretical insights and methodological references for the application of SSVEP-based brain–computer interfaces in real-world scenarios.

Motor imagery (MI) is a popular noninvasive brain computer interface (BCI) paradigm, yet its decoding accuracy remains hindered by the inherent nonstationarity and low signal-to-noise ratio of electroencephalogram (EEG) signals. Current decoding frameworks often fail to fully exploit the intricate spatial-temporal dependencies, leading to suboptimal feature representation and the omission of latent discriminative cues. To address these challenges, we introduce a deep neural network-powered multifaceted strategy (DPMS-Net) model, a novel approach that employs dynamic convolution to unearth effective discriminative cues across multiple dimensions, including the temporal, spatial, and frequency domains. This model synergizes channel and temporal attention mechanisms to adeptly capture the salient features of EEG signals across diverse spatial-temporal dimensions, thereby mitigating the risk of omitting critical information. Furthermore, we introduce a spectral-domain analysis component that unearths subtle oscillatory signatures hidden within the EEG spectrum, providing enriched evidence for classification. We evaluated the performance of DPMS-Net on two publicly available datasets and a self-collected dataset from stroke patients. On the BCI Competition IV 2a and BCI Competition IV 2b datasets, DPMS-Net achieved subject-dependent classification accuracies of 83.93% and 88.38%, respectively, alongside subject-independent classification accuracies of 65.88% and 76.01%. In the stroke patient dataset, DPMS-Net attained a subject-dependent classification accuracy of 67.67% and a subject-independent classification accuracy of 57.58%. Experimental results indicate that DPMS-Net possesses efficient decoding capabilities and robust stability, reflecting its potential for deployment in neurorehabilitation BCI systems.

Objective Motor learning is key to successful neuro-rehabilitation. Combinations of Brain-Computer Interfaces (BCIs) and repetitive transcranial magnetic stimulation (rTMS) have been proposed for neurorehabilitation following conditions such as stroke. However, rTMS is typically delivered via a fixed protocol without taking into consideration the current brain states of participants. We propose a new BCI-based rTMS delivery protocol for supporting motor learning. Specifically, we propose BCI-based brain state dependent delivery of rTMS, in which a BCI system measures the event-related desynchronisation (\ERD; a neural marker of motor learning in the alpha band, selected because it is a robust, well-established real-time EEG correlate of motor activity and cortical excitability) in order to determine when to deliver rTMS. Approach We compare our proposed rTMS delivery protocol with two state of the art comparable protocols: delivery of rTMS prior to the BCI-based motor learning and delivery of rTMS at fixed times throughout the experiment, as well as a control condition in which no rTMS was used. Each protocol is tested with a different group (n=8) of participants (n=32 total participants). Main Results Our results reveal a significant effect of changing the rTMS delivery protocol ($p=0.005$) and that our proposed rTMS delivery protocol delivers better motor learning outcomes than other state of the art rTMS delivery protocols (e.g. BCI group vs. fixed times group: p=0.003, BCI group vs. no rTMS group: p=0.03). Inspection of ERD dynamics from each of our participant groups demonstrates that our BCI-based rTMS paradigm keeps corticospinal excitability relatively stable throughout the learning period, keeping the brain in a more optimal learning state for longer. Significance These findings suggest potential applications for adaptive rTMS–BCI systems in clinical neurorehabilitation, sports skill learning, and neuroprosthetic control.