Decoding Algorithms

Brain-computer interface decoders can guess the right label using information unrelated to the target mental state. A team at the School of Psychology, Beijing Language and Culture University (BLCU), proposes a falsifiable substitution-test standard: candidate evidence must persist in disjoint data, survive capacity-matched substitutions of physical organization or listener templates, and remain testable after target events are excluded. Across six EEG datasets (41 participants), averaging four neural-speech margin metrics brought 5-second decoding to what the authors call a leading level; in two hierarchical interfaces, parent-stream error scores kept AUCs of 0.968 and 0.965 after all target-command events were excluded. The framework offers a test for attributing evidence in neuroscience and BCI.

Researchers from Southwest State University developed an integrated brain-computer interface system supporting both synchronous and asynchronous paradigms, achieving 60-65% recognition accuracy in EEG-based robotic device control tasks.

Researchers propose ReCIL, a rehearsal-based class incremental learning method for cross-subject motor imagery classification that lets a model learn new MI classes sequentially without retraining from scratch. Using Euclidean alignment to reduce cross-subject EEG distribution shift and global-local replay to preserve earlier-task knowledge, ReCIL achieved a good balance between plasticity and stability across three public MI datasets. The authors report it as the first study of cross-subject class incremental learning for MI classification.

The study proposes BiGSTF-Net, a multimodal architecture that exploits the complementary properties of EEG and functional near-infrared spectroscopy (fNIRS) for cognitive-state decoding: heterogeneous spatio-temporal extractors capture each modality's representations, a modal residual interaction unit provides bidirectional cross-modal guidance, and a spatio-temporal gating unit fuses intra-modal features. Under cross-session evaluation on multiple BCI datasets, BiGSTF-Net consistently outperformed representative multimodal fusion baselines; ablations validated each component and visualizations matched the known neurophysiological features of the two signals.

A preprint introduces NeuroPB, a framework that scales neural decoding by transferring pretrained behavioral representations: a motion encoder is first pretrained on large-scale motor behavior data, then limited paired neural-behavior recordings align neural activity to the behavioral representation space, and a neural encoder with a lightweight motor decoder reconstructs continuous movement. The researchers report that behavioral pretraining lifted center-out trajectory decoding R² by 11% and random-target task performance by 8%, that pretraining on robotic trajectories matched macaque-trajectory performance, and that only 10% calibration was needed to match training from scratch. The study is a preprint and has not been peer-reviewed.

A preprint introduces OSPDIM, a source-free online unsupervised domain adaptation framework for class-imbalanced label shift in online EEG brain-computer interfaces. The study argues that Riemannian alignment methods such as the Riemannian centering transform handle covariate shift but implicitly assume balanced class priors, so shifting label distributions in real online use geometrically misalign the target data; OSPDIM adds a manifold-constrained bias parameter in the Riemannian tangent-space mapping, optimized by information maximization, and estimates and corrects the geometric bias online rather than relying on global batch statistics. 2D SPD matrix simulations showed OSPDIM correcting misalignment that standard centering fails on, and across multiple motor imagery datasets it significantly outperformed standard Riemannian baselines, especially online. The study is a preprint and has not been peer-reviewed.

The study presents an EEG brain-computer interface signal-processing framework that combines hybrid MothCray channel-selection optimization — fusing moth-flame and crayfish optimization — with deep learning. After notch filtering, independent component analysis (ICA) and time-window segmentation, the MothCray algorithm identifies the most informative channels and a deep neural network adapted to EEG spatio-temporal features classifies. The model reached 93.92% accuracy on BCI Competition IV dataset IIa, ahead of existing methods.

The study presents AutoMI, a framework that uses LLM multi-agents to automatically and rapidly iterate on motor imagery EEG classification models, combining a Q-learning policy with deterministic rules and integrating planning, execution and output agents with predefined tools, plus experience tracking and rollback. Models built by AutoMI reached 77.62%, 78.08% and 83.02% accuracy on the IV2a, OpenBMI and ECUST-MI datasets — up 18.42%, 9.27% and 19.25% over automated optimization algorithms.

Researchers report FRED, a principled EEG-fNIRS fusion framework for imagined handwriting decoding, posted to arXiv as a preprint on August 4, 2026 and not yet peer reviewed. Imagined handwriting offers a temporally rich paradigm for non-invasive neural decoding, and EEG-fNIRS fusion promises complementary neural information, but fusion is typically heuristic and lacks principled treatment of frequency-band redundancy. FRED builds frequency-decorrelated temporal ensembles for imagined handwriting decoding. The ensemble reaches 0.8076/0.7242/0.7492 accuracy on the public/private/overall test partitions without test-set adaptation or output constraints, and the complete system reaches 0.8498/0.7718/0.7952, ranking fourth on the private split. A modality audit finds fNIRS-only decoding at chance (0.2511 overall), while adding fNIRS to EEG changes accuracy by only +0.0025.

The study recorded hippocampal CA1 population activity in rats performing a sequential spatial decision task in a modified T-maze, dividing the decision into initiation, running and approach phases. Local theta sequences consistently over-represented the actual choice, while prospective representations driven by choice-arm place cells shifted from predicting the actual choice during running to representing potential paths more evenly at the choice point. Integrating local and prospective features improved decoding, reaching 74.4% accuracy for future choice prediction and 78.2% for upcoming trajectory decoding.

A preprint presents STEAM, a hierarchical transfer framework for EEG decoding that pairs a dual-branch spatio-temporal encoder with a shared soft mixture-of-experts (SSMoE) module to reconcile general representation learning with paradigm-specific specialization in EEG foundation models, letting complementary representations exchange information through a compact set of soft slots. The researchers report that STEAM attained the best average rank at competitive inference cost across 7 downstream datasets and 14 evaluation settings, and that its hierarchical pretraining further lifted decoding accuracy without retraining from scratch. The study is a preprint and has not been peer-reviewed.

Baylor College of Medicine researchers analyzed more than 200 hours of intracranial recordings from eight patients with treatment-resistant obsessive-compulsive disorder, five of whom also had bilateral orbitofrontal ECoG electrodes. Neural signals alone did not decode symptom severity or momentary distress above chance; adding facial and speech features raised performance to R=0.73 for severity and R=0.34 for distress. The medRxiv preprint has not been peer reviewed.

CORTIVA, a candidate-score fusion framework for EEG- and MEG-to-image retrieval, reached 73.5% Top-1 and 95.3% Top-5 accuracy across ten participants on the 200-way THINGS-EEG2 benchmark, beating the strongest reported baseline by 10.3 and 5.4 percentage points. Instead of compressing heterogeneous visual supervision into a single embedding before ranking, the authors let separate decoding routes align to different visual targets and score the same candidate pool independently, merging only the temperature-scaled score vectors, which they say preserves complementary evidence; with a modality-specific encoder the same approach reached 42.4% Top-1 on THINGS-MEG. The work is a preprint and has not been peer reviewed.

The study presents a spatial harmonization framework built on a two-layer graph convolutional network (GCN) that maps heterogeneous EEG recordings onto a unified physical electrode layout while preserving motor imagery information, addressing electrode-configuration mismatches across MI-BCI datasets. Each trial is modeled as a graph so the GCN captures spatio-temporal relations, and harmonized EEG showed lower error than spherical spline interpolation while retaining the key temporal-spectral-spatial features. Combining real and harmonized EEG lifted EEGNet accuracy from 56.57% to 66.20% and FBCNet from 61.96% to 72.54% in within-session classification on Dataset A, and supported source-only cross-dataset transfer and target-domain fine-tuning.

This preprint introduces EasyBCI, which the authors say automates BCI preprocessing across six signal types with a two-phase large language model agent. The study is a preprint and has not been peer reviewed. On EEG with a fixed linear classifier, the authors report that all five EasyBCI backbones preserve more task-relevant separability than the manual pipeline, and that the system extends to five additional modalities spanning nearly three orders of magnitude in sampling rate.

A study of 60 patients with subacute stroke hemiplegia found that adding motor imagery brain-computer interface training to conventional rehabilitation significantly improved upper limb motor function, simplified upper limb function scores, and daily living abilities. The experimental group of 30 received 3 weeks of additional BCI training, 5 days per week, while the control group received only conventional rehabilitation. The experimental group showed greater improvements in Fugl-Meyer upper limb scores, simplified upper limb function scores, and Barthel Index, with statistically significant differences.

Researchers at the University of Michigan, Ann Arbor have built SpikeCleaner, an algorithm that grades neural units after automated spike sorting, reaching 97% accuracy and a 92% F1 score on single units in benchmarking. It combines spike rate, spike-timing metrics and waveform features to classify units as good, noise or multi-unit activity, a job that otherwise falls to manual curation.

Researchers in Pakistan and South Korea have built EmoWrite, a brain-computer interface that uses sentiment analysis to turn thought into text, typing at 6.6 words per minute with 90.36% accuracy in tests with 72 volunteers. The system adds a dynamic keyboard and visual feedback to make the interface easier to use.

Researchers writing in npj Biomedical Innovations have proposed DSTF-Net, a framework that decodes steady-state visual evoked potentials (SSVEP) from frontal EEG alone, removing the need for electrodes over the occipital cortex. In cross-subject transfer to 20 new users, including eight brain-injured patients lying supine, it improved decoding accuracy by as much as 33.47% over baseline methods.

Brown University researchers have posted a preprint proposing Neural SS-DMP, a movement decoder that does not output hand coordinates directly: it first infers a compact set of parameters describing the motion the user intends, then hands them to a generator governed by physical dynamics that draws the full trajectory, so decoded output stays within motion a body can actually produce. The generator is tuned per person, blending general movement dynamics with the individual's own patterns estimated from training data, and the authors say that across two kinds of neural recording the model came out ahead of strong existing methods on both accuracy and trajectory smoothness while holding performance longer on recordings made after training ended. The study is a preprint, has not been peer reviewed, and its results come entirely from offline data rather than live control.

A preprint proposes UnSPC (uncertainty-guided self-paced cyclic learning), a framework that integrates domain adaptation (DA) and domain generalization (DG) in an iterative cycle to address neural drift in long-term invasive brain-machine interfaces, where drift erodes decoding performance and forces frequent recalibration and where existing methods rely on DA or DG alone. Under an uncertainty-guided, self-paced pseudo-labeling scheme with a noise-robust ranking strategy, UnSPC iteratively mines reliable pseudo-labeled samples and, through cyclic adaptation and generalization, gradually mitigates both global and sub-domain drift. Experiments on multiple neural decoding datasets validated its effectiveness and robustness; the authors say it is the first method to integrate DA and DG through a pseudo-label loop. The study has not been peer reviewed.

The VA Office of Research and Development is advancing a high-performance implanted brain-computer interface (BCI) to improve independence for Veterans and others with tetraplegia or inability to speak due to ALS, spinal cord injury, or stroke. The project enhances deep learning decoders and multi-state gesture decoding, deployed on a battery-powered mobile BCI device for independent home use of computers and touch-enabled devices. Accuracy and usability will be evaluated in participants already enrolled in the BrainGate investigational clinical trial.

UC Davis researchers trained a transformer-based speech decoder across six people with intracortical BCIs. The pooled model cut relative word error rates by more than 50% on average compared with subject-only models and, after fine-tuning on fewer than 200 sentences from an unseen user, achieved a word error rate below 7%. The bioRxiv preprint has not been peer reviewed.

Researchers paired interquartile-range clipping for outlier removal with a random forest classifier on the UCI machine learning repository's EEG-Eye-State dataset, classifying eyes-open versus eyes-closed at 92.49% accuracy with a ROC-AUC of 0.9791. Cross-validation put mean accuracy at 92.86%. The authors present the pipeline as a stable, interpretable option for BCI uses such as drowsiness monitoring and assistive technology.

A preprint proposes SSCDL, a self-supervised consistency-enhanced disentangled learning framework that decomposes motor signals into velocity, direction, and speed to capture representations invariant to neural drift and significantly enhance cross-day decoding generalization for invasive BMIs. The researchers report state-of-the-art decoding performance with high robustness and cross-day stability across extensive experiments; the work has not yet been peer reviewed.

UC Irvine researchers tested an EEG category-decoding task in 30 participants viewing pictures and words from five semantic groups. All picture-category pairs were statistically separable, but only one word-category pair was; parietal and left-temporal electrodes contributed more to picture decoding than frontal and right-temporal sites. The bioRxiv preprint has not been peer reviewed.

University of British Columbia researchers released BEND-BCI, an open benchmark comparing 23 neural decoders across 16 real or synthetic recordings in motor, visual, speech and spatial tasks. Rankings often changed when robustness, computational cost and cross-recording representation consistency were considered alongside held-out accuracy, and simpler baselines sometimes matched or beat much larger deep networks. The bioRxiv preprint has not been peer reviewed.

A preprint proposes an adaptive EEG generation framework based on conditional flow matching to ease data scarcity in brain-computer interfaces and support large-scale neural modeling. Noting that existing flow methods assume one global time course across all channels and time segments, the framework adds position-adaptive time scheduling that tracks per-position reconstruction error to modulate each position's time course, plus decomposed spatiotemporal attention and a frequency-aligned multi-resolution spectral consistency loss to model cross-channel dependencies and compensate for EEG's power-law spectral bias. Across three EEG datasets with different acquisition protocols and task semantics, it consistently beat the strongest baselines, cutting TS-FID by up to 62.2% and lifting downstream classification accuracy by up to 6.77 percentage points. The study has not been peer reviewed.

Researchers report a preprint, posted to arXiv on July 25, 2026 and not yet peer reviewed, that contrasts Bayesian complete-pooling models against frequentist baselines for cross-subject, left-hand versus right-hand motor imagery EEG classification across 20 datasets. Six frequentist pipelines were each paired with an analogous Bayesian pipeline sharing identical feature engineering and fit via Markov chain Monte Carlo posterior sampling. Bayesian complete-pooling produced statistically but not practically significant improvements in reliability and increased predictive uncertainty, with no significant differences in Brier score, resolution, or discrimination. Bayesian pipelines consumed roughly 13 times more energy than their frequentist counterparts, and the authors conclude that complete pooling alone offers limited practical benefit, pointing to partial pooling as a more promising direction.

The researchers developed a deep learning architecture that analyzes multichannel interictal intracranial EEG to localize epileptogenic zones without requiring seizure-period recordings or manual annotation. The model combines a Morlet-wavelet temporal Transformer with a spatial attention encoder. It was evaluated with 50.5 hours of recordings from 161 patients and 17,012 channels at 7 independent centers. Leave-one-center-out validation produced a pooled AUROC of 0.778 with a 95% confidence interval of 0.748 to 0.808, and discrimination remained above chance at every held-out center. The results indicate performance comparable to established electrophysiological baselines across centers and implantation modalities, but prospective clinical validation is still needed. This medRxiv preprint has not been peer reviewed.