The choice of R as the implementation language reflects both practical and cultural considerations within the statistical and neuroimaging communities. R remains one of the most widely used environments for behavioral and biomedical data analysis, and its package ecosystem provides established mechanisms for documentation, testing, and distribution through repositories such as CRAN and GitHub. Versioned releases are particularly consequential for longitudinal studies like ABCD, where data are re-released periodically and analyses must remain anchored to specific dataset versions. Under the previous fragmented approach, a scoring algorithm could change between releases without any formal record, making it difficult to determine whether differences in findings across papers reflected true developmental change or shifts in scoring conventions. With ABCDscores, each release of the package corresponds to a defined set of algorithms, and the source code history on version-control platforms documents every modification, creating a parallel versioning system that researchers can align with the tabulated data releases they download.

The modular architecture described by the authors deserves particular attention because it addresses a common failure mode in large collaborative software projects: the accumulation of idiosyncratic, one-off scripts that resist maintenance. By requiring every scoring function to follow a uniform naming convention and a consistent input-output contract, the package lowers the cognitive burden on both contributors and users. A researcher who understands how one score is computed can reasonably infer how any other score is computed, since the pattern repeats across the codebase. The organization of functions by research domain—general, family and community, mental health, neurocognition, novel technologies, physical health, and substance use—mirrors the structure of the tabulated dataset itself, so the mapping between a data column and the code that generated it is largely self-evident from the naming scheme. This alignment between data structure and code structure is a subtle but important design decision, as it reduces the likelihood that users will apply the wrong scoring function to the wrong set of items.

The inclusion of companion variables such as the number-missing indicators illustrates a broader principle in psychometric scoring that is often underappreciated by analysts new to questionnaire data. Most established instruments were normed under conditions in which all items were administered, and the validity of a summed or averaged score can degrade when items are missing, particularly if the missingness is not random. By distributing _nm variables alongside every mean and sum, the package gives analysts the information needed to apply sensible exclusion criteria or to model missingness explicitly rather than treating all computed scores as equally trustworthy. This kind of transparency about data quality at the item level is especially valuable in a longitudinal cohort, where attrition and intermittent nonresponse accumulate over successive waves and can interact with the constructs being measured.

The separation of particularly complex scoring procedures into dedicated source files also signals where the greatest methodological risk historically lay. Instruments such as the Achenbach System of Empirically Based Assessment involve elaborate rules for handling skipped items, partial responses, and age-dependent norms, while substance-use timeline follow-back procedures require careful calendar-based reconstruction of self-reported behavior. Family history and medication data introduce their own complications, including coded narratives and changing formulations over time. Encapsulating these procedures in well-defined functions with explicit logic means that methodological debates about how such scores should be computed can now occur in the open, with reference to specific lines of code rather than to undocumented conventions known only to the study’s internal data team.

From the perspective of the broader open-science movement, ABCDscores follows a trajectory established by earlier large-scale initiatives. The UK Biobank and the Human Connectome Project demonstrated that sharing de-identified data at scale could accelerate discovery across thousands of laboratories, but both also revealed that data sharing alone is insufficient: derived variables must be shared with equal care, or the community reproduces them inconsistently. The ABCD Study’s decision to treat its scoring software as a first-class scientific product, rather than as internal plumbing, represents a maturation of this model. It acknowledges that in modern computational science, the code that generates a variable is as much a part of the methods section as the description of the measurement instrument itself.

The practical benefits for the research community extend beyond reproducibility into efficiency and error reduction. Recomputing summary scores from raw items has historically required researchers to reverse-engineer scoring rules from instrument manuals, a process that is time-consuming and error-prone, particularly for instruments with conditional branching or reverse-coded items. By providing vetted implementations, the package allows investigators to redirect effort from reimplementation toward substantive scientific questions. It also reduces a source of between-lab variability: when two research groups compute the same construct using independently written scripts, small discrepancies in handling of edge cases can produce different values for the same participant, complicating meta-analysis and replication. A shared reference implementation eliminates this class of discrepancy at its origin.

The customization capability emphasized by the developers also merits consideration. Standardized scores serve the majority of users, but a substantial fraction of ABCD analyses require modified scoring—for example, alternative subscale aggregations, item-level factor models, or adaptations for special populations. Because the package exposes its scoring logic as ordinary R functions operating on data frames, researchers can compose, extend, or override individual steps without forking the entire pipeline. This design keeps the boundary between official and custom scoring explicit: any deviation from the standard implementation is visible in the analyst’s own code, rather than buried in undocumented modifications. Such clarity helps maintain the distinction between consensus measures used for comparability across studies and exploratory measures developed for specific hypotheses.

There is also an institutional dimension to this work. The ABCD Study involves 21 research sites, and coordinating data collection across such a distributed network has always required careful harmonization of administration procedures. Harmonization of derived variables is an equally demanding task, and the shift from REDCap-calculated fields and scattered external scripts to a single maintained package effectively centralizes that responsibility in a reviewable artifact. When a scoring bug is discovered, the fix now needs to be made and validated in one place, and every subsequent release inherits the correction, rather than requiring parallel patches across multiple systems and programming languages. This consolidation reduces both the probability of errors and the cost of correcting them when they occur.

Looking forward, the authors position ABCDscores as a potential template for other longitudinal cohort studies, and the design choices they describe suggest why. The combination of domain-based organization, consistent function contracts, versioned releases, and open source code constitutes a reusable pattern rather than a one-off solution. As population neuroimaging and developmental cohorts continue to grow in number and scale, the bottleneck in many analyses is shifting from data acquisition to data preparation, and tools that make preparation transparent and reliable will increasingly determine the pace and trustworthiness of cumulative science. For the thousands of researchers who will analyze ABCD data in the coming years, the package transforms summary scores from a black box into a shared, inspectable foundation—one that can evolve with the study itself as new waves of data, new instruments, and new scoring standards emerge over the cohort’s journey from middle childhood into adulthood.

Subject of Research: Neuroinformatics

Article Title: ABCDscores: an R package for computing summary scores in the ABCD Study®

Article References: Zhang, L., Celhay, O., Das, B., Berman, S., Ziemer, L. R., Smith, C. J., Abad, S., Dale, A. M., & Linkersdörfer, J. (2026). ABCDscores: an R package for computing summary scores in the ABCD Study®. Frontiers in Neuroinformatics, 20, Article 1858950. https://doi.org/10.3389/fninf.2026.1858950

Image Credits: AI Generated

DOI: 10.3389/fninf.2026.1858950

Keywords: ABCD Study, R package, neuroinformatics, reproducibility, brain development, open science, ABCDscores, package, computing, summary, scores, ABCD

The central premise of the proposal is that a useful digital framework in autism should be construct-preserving rather than construct-erasing. Many contemporary AI systems optimize prediction directly from raw data without retaining human-interpretable inferential layers. That approach may be appropriate for some tasks, but the author argues it is poorly suited to a clinical ecosystem in which the meaning of a measurement emerges from the interaction of several inferential layers rather than from raw signals alone. FIAP®-Digital is therefore designed to integrate multimodal inputs, including biological, physiological, developmental, contextual, therapeutic-process, and longitudinal indicators, into structured estimates of clinically defined constructs such as the Biological Burden Index, energetic capacity, adaptive neurodevelopmental window accessibility, therapeutic engagement accessibility, neuroplastic accessibility, and higher-order translational profiles.

Technically, the proposed architecture is organized into six computational layers. The first is a multimodal input architecture that organizes heterogeneous data streams by modality and temporal structure, ranging from biomarkers and physiological measures to developmental and therapeutic-process variables. The second is a construct-estimation engine that converts raw data into interpretable translational layers such as the Biological Burden Index and energetic capacity. The third is a translational profile-inference engine that infers higher-order multidomain configurations, including burden-dominant, reserve-constrained, timing-sensitive, engagement-vulnerable, and mixed profiles, from the interaction among constructs rather than from raw data directly. The remaining layers handle explainability and clinician-in-the-loop review, precision-oriented research outputs, and longitudinal updating that tracks how profiles shift across time.

The author identifies a specific architectural gap in existing autism AI research. Many machine learning approaches are optimized for diagnostic classification, biomarker discovery, behavioral phenotyping, or outcome prediction, and multimodal data are often fused into statistical representations that may improve performance while reducing visibility of the intermediate clinical and biological constructs that matter for interpretation. Few computational frameworks explicitly connect biological burden, adaptive reserve, developmental timing, therapeutic engagement, neuroplastic accessibility, and intervention responsiveness. FIAP®-Digital differs from generic multimodal precision psychiatry pipelines along six dimensions: its primary goal is construct-guided stratification rather than diagnostic classification; its intermediate representations are explicitly defined constructs rather than opaque latent variables; its outputs are clinician-readable research profiles rather than autonomous decisions; its explainability is built around construct contributions, missingness, and uncertainty rather than post-hoc feature importance alone; its validation roadmap includes construct validity, usability, fairness, and uncertainty calibration rather than model performance alone; and its current context of use is hypothesis generation and prototype testing, not clinical decision support.

On the question of fusion, the paper argues that intermediate or construct-level fusion is likely the most appropriate starting point because it preserves interpretability. Rather than merging all data into a single undifferentiated feature space, modality-specific indicators would first be mapped to candidate constructs. Sleep disruption, gastrointestinal burden, fatigue, and stress reactivity, for example, would be integrated into a Biological Burden Index-related representation, while regulation, attention, persistence, and responsiveness to support would be integrated into a therapeutic engagement representation. Higher-order translational profiles would then be inferred from interactions among construct-level estimates. The objective is not only predictive accuracy but preservation of translational meaning, a distinction the author argues separates the architecture from purely performance-driven multimodal systems.

Temporal modeling is a core requirement because autism intervention responsiveness is dynamic. Biological burden, energetic reserve, engagement availability, contextual demands, and developmental readiness may vary across time, so the architecture requires longitudinal updating rather than static one-time classification. Candidate approaches include longitudinal mixed-effects models, state-space models, dynamic Bayesian updating, repeated-measures trajectory models, temporal latent profile models, recurrent architectures, or transformer-based longitudinal representations. In early feasibility phases, the paper suggests that simpler longitudinal tracking and rule-based temporal summaries may be more appropriate than complex deep learning models, with gradual methodological escalation as data volume and validation maturity increase.

Missing data and uncertainty receive equally prominent treatment. Missingness in autism research is not treated merely as a technical nuisance; it may carry interpretive meaning when certain modalities are systematically unavailable because of service access, family burden, sensory intolerance, cost, geographic location, or institutional constraints. The architecture would include modality-specific missingness flags, data-completeness scores, and uncertainty inflation when key information is unavailable. Uncertainty would be visible at multiple levels, from variable to construct to profile to longitudinal interpretation. A burden-dominant profile might be estimated with high confidence when multiple convergent indicators are present, but with low confidence when key biological or contextual data are missing. When data are insufficient to estimate a construct reliably, the system should explicitly indicate low confidence rather than generating a deterministic profile.

Human supervision is positioned as the governance layer that makes AI use proportionate and ethically bounded. AI is assigned six restricted functions: organizing heterogeneous inputs by modality, source, timestamp, quality, and missingness; harmonizing variables across records and sites; mapping indicators to provisional FIAP® constructs; estimating construct-level profiles with uncertainty flags; identifying longitudinal patterns and profile shifts; and generating clinician-readable explanations of profile drivers, data gaps, and uncertainty. The system does not make autonomous diagnostic, prognostic, eligibility, or treatment decisions. Clinicians and qualified researchers remain responsible for interpretation, contextualization, contestability, and deciding whether an output is meaningful for research use. The paper also proposes a staged validation roadmap covering feasibility testing, construct validity, internal and external validation, longitudinal robustness, uncertainty calibration, clinician usability, fairness assessment, and low-resource adaptation, with a tiered architecture that begins with clinically accessible information such as caregiver report and clinician observation before adding biomarkers, physiological sensors, or video analysis only when feasible and ethically justified.

The boundaries of the proposal are stated with unusual clarity. FIAP®-Digital is not presented as a validated diagnostic tool, treatment recommendation system, autonomous AI model, or clinically proven platform. It should not be used to diagnose autism, assign treatment, determine service eligibility, predict individual outcomes, or replace clinical judgment. The core FIAP® constructs remain hypothetical and have not undergone formal construct validation, and no clinical evidence of effectiveness is claimed. The manuscript is a conceptual article that reports no original empirical data, prototype testing, computational experiments, or clinical trial findings, and the proposed computational architecture is not an implemented algorithm. The author acknowledges that some constructs may require revision, subdivision, or replacement after empirical testing.

The primary contribution, then, is to define a responsible computational pathway for testing whether FIAP® constructs can support future precision-stratified autism research. If the constructs survive empirical evaluation, the framework may eventually contribute to precision-stratified autism research and to more responsible, interpretable, and developmentally sensitive digital phenotyping approaches. Until then, the architecture stands as a neuroinformatics foundation and validation roadmap, specifying how future prototypes could be designed, audited, and empirically tested before any clinical translation is considered.

Subject of Research: Neuroinformatics

Article Title: Toward FIAP®-Digital: an interpretable multimodal AI architecture for translational stratification and precision care in Autism

Article References: Fuamba, Y. (2026). Toward FIAP®-Digital: an interpretable multimodal AI architecture for translational stratification and precision care in Autism. Frontiers in Neuroinformatics, 20, Article 1868068. https://doi.org/10.3389/fninf.2026.1868068

Image Credits: AI Generated

DOI: 10.3389/fninf.2026.1868068

Keywords: artificial intelligence, autism spectrum disorder, explainable AI, multimodal integration, neuroinformatics, precision-stratified research, responsible AI, translational stratification, Toward, FIAP, Digital, interpretable

The clinical motivation behind automated seizure detection rests on a simple arithmetic problem. A single overnight EEG recording can span eight or more hours, sampled continuously across dozens of electrode channels, yet a typical seizure may occupy only a few minutes of that entire span. Trained electroencephalographers must therefore scan vast stretches of unremarkable activity to find brief pathological events, and their performance can degrade with fatigue, particularly during extended overnight shifts. The burden is compounded by the fact that seizure morphology differs substantially between patients and even between events in the same patient, so pattern recognition cannot rely on a single stereotyped signature. Automated systems that pre-screen recordings and flag candidate events allow human experts to concentrate their attention on the segments most likely to contain abnormal activity, transforming a review task measured in hours into one measured in minutes.

The bipolar montage that MGF-Net exploits as a structural prior is itself a product of decades of clinical convention. In bipolar recording, each displayed channel represents the voltage difference between two neighboring electrodes rather than the voltage at a single electrode relative to a distant reference. This arrangement has a practical advantage: activity common to both electrodes in a pair, such as widespread artifact or distant noise, is largely cancelled by the subtraction, while activity localized near one electrode is preserved and even emphasized. Clinical readers typically inspect recordings arranged in chains of such electrode pairs, for example running anterior to posterior across each hemisphere, because this organization makes it visually apparent when a discharge propagates along the cortical surface. The 18 fixed derivations used in the study reflect this standard arrangement, and the relationships among them, adjacency within a chain, mirror positions across hemispheres, and shared electrodes between chains, are not arbitrary but follow directly from how the electrode pairs are constructed.

What distinguishes the graph-based formulation adopted here from earlier multichannel approaches is the typing of these relationships. In a conventional convolutional network, information from all channels is blended within shared feature maps, and the network must implicitly discover any spatial organization from data alone. Generic attention mechanisms likewise treat channels as an unordered set unless positional information is injected. Graph neural networks offer an intermediate solution by specifying which nodes may exchange information, but many prior applications have used a single undifferentiated connectivity pattern, so the model cannot distinguish a neighbor along the same montage chain from an electrode on the opposite side of the head. By labeling each edge with its relation type and assigning relation-specific projections within the attention mechanism, MGF-Net allows the network to learn different interaction patterns for each kind of relationship. A derivation can, in effect, ask different questions of its same-chain neighbor, its contralateral counterpart, and the derivation with which it shares an electrode.

The signal preprocessing pipeline also reflects practical constraints of long-term monitoring. Epoch-wise robust normalization, applied independently to each short segment of the recording, guards against the slow drifts and variable impedance conditions that afflict scalp electrodes over many hours of continuous wear. Wavelet-based signal reconstruction provides a means of suppressing components that fall outside the frequency bands of physiological interest, since seizure rhythms typically occupy well-characterized ranges while muscle artifact, electrode pops, and environmental interference can contaminate other bands. Applying these steps per epoch rather than per recording means the system can adapt to non-stationary signal statistics without requiring knowledge of the full recording in advance, which matters for prospective deployment where detection must proceed as data streams in.

The choice of evaluation metrics in this study separates two distinct questions that are often conflated. Segment-level performance, reported here as 98.38 percent accuracy with 93.18 percent sensitivity and 98.41 percent specificity, measures how well the model classifies individual short time windows as ictal or non-ictal. Event-level performance, by contrast, asks whether the system correctly identifies discrete seizure episodes as clinical entities, merging consecutive positive segments into single detections. A model can achieve respectable segment-level statistics while producing fragmented or duplicated event detections, so event-level evaluation with false detection rate and latency measures is closer to what matters clinically. The reported false detection rate of 0.82 per hour means that over a typical overnight recording, a reviewer would need to dismiss fewer than seven false alarms, a manageable burden compared with unassisted review. The 2.03-second average latency indicates that detections arrive essentially in real time, which opens the possibility of alerting clinical staff during the recording rather than only afterward.

The CHB-MIT dataset on which the system was evaluated deserves some context. It is a publicly available collection of pediatric scalp EEG recordings from patients with medically refractory epilepsy, assembled at Children’s Hospital Boston, and it has become the de facto benchmark for seizure detection research. Its widespread use enables direct comparison between methods, but it also carries limitations: the patient population is young, the recordings derive from a single institution with a consistent acquisition protocol, and the annotation standards reflect the practices of one monitoring unit. Performance on CHB-MIT does not automatically transfer to adult populations, to recordings acquired with different electrode systems, or to seizure types that are underrepresented in the dataset. The authors’ use of a fixed set of 18 bipolar derivations aligns with the standard 10-20 electrode placement used in the dataset, but clinical recordings elsewhere may include additional electrodes or different montage conventions that would require adaptation.

The architecture’s computational design also merits attention. Sharing a single one-dimensional convolutional network across all 18 derivations serves two purposes: it drastically reduces the number of parameters compared with per-channel encoders, and it forces all derivations into a common feature space so that subsequent graph operations compare like with like. The relation-aware graph Transformer then performs the cross-derivation fusion, and attention-based pooling compresses the 18 updated node representations into a single vector per epoch. This pooling is learned rather than fixed, meaning the network can weight derivations differently depending on the content of each epoch, potentially attending more closely to the channels near a seizure focus while discounting distant or artifact-laden derivations. Such learned selectivity mirrors, in a rudimentary way, what experienced readers do when they focus on the regions of the montage most relevant to a suspected event.

Post-processing constitutes the final stage that converts raw epoch probabilities into clinically usable event detections. Epoch-wise classifiers inevitably produce scattered false positives, isolated windows flagged amid normal activity, and brief gaps within genuine seizures. Smoothing and thresholding operations over sequences of probabilities consolidate these into coherent event boundaries. The details of this stage, though less glamorous than the neural architecture itself, materially affect the event-level metrics that clinicians care about, and the favorable false detection rate reported here suggests the post-processing was tuned with that trade-off in mind.

Looking forward, the montage-as-graph framing may generalize beyond seizure detection. The same structural prior, adjacency, bilateral correspondence, and shared electrodes, exists in any bipolar EEG analysis, including sleep staging, encephalopathy monitoring, and intraoperative neuromonitoring. Relation-labeled graphs could also accommodate variable montages by relabeling edges rather than retraining the architecture. More broadly, the study illustrates a growing trend in biomedical machine learning: rather than treating domain structure as an obstacle to be learned from scratch, encoding established clinical conventions, such as how electroencephalographers organize and compare channels, directly into the model’s inductive biases can yield systems that are both more accurate and more interpretable to the practitioners who must ultimately trust their output.

Subject of Research: A graph Transformer network that uses bipolar EEG montage structure to detect seizures in long-term scalp EEG recordings

Article Title: Montage-guided fusion network for long-term scalp electroencephalography seizure detection

Article References: Yin, H., & Yin, X. (2026). Montage-guided fusion network for long-term scalp electroencephalography seizure detection. Frontiers in Neuroinformatics, 20, Article 1942480. https://doi.org/10.3389/fninf.2026.1942480

Image Credits: AI Generated

DOI: 10.3389/fninf.2026.1942480

Keywords: epilepsy, seizure detection, scalp EEG, graph Transformer, bipolar montage, deep learning, CHB-MIT, neuroinformatics, wavelet transform, attention pooling, clinical neurology, machine learning

The appeal of EEG as a screening tool is obvious. It is cheap, non-invasive, widely available, and offers excellent temporal resolution, recording the functional changes in the cerebral cortex that reflect neuronal degeneration even before clear behavioral symptoms appear. Decades of research have established three characteristic EEG signatures of Alzheimer’s disease: diffuse slowing, in which spectral power decreases in higher frequency bands (alpha, beta, gamma) and increases in lower bands (delta, theta); reduced signal complexity, measurable with entropy-based metrics; and disrupted functional connectivity between cortical areas, signaling the breakdown of large-scale brain networks. These neurophysiological changes are thought to arise from cholinergic dysfunction, cortical hyperexcitability, and neocortical disconnection. The problem has never been whether the information is in the signal—it is whether machine learning models can extract it reliably, interpretably, and across different recording hardware.

That is where existing approaches have fallen short. Traditional methods rely on hand-crafted features—spectral power, entropy, connectivity measures—fed into classifiers such as support vector machines and random forests. These pipelines cannot capture long-range temporal dependencies in resting-state EEG, tend to overfit on small single-dataset cohorts, and perform inconsistently across different EEG systems and montages. Deep learning architectures such as EEGNet, EEGInception, and EEGConformer learn features end-to-end, and transformer-based foundation models like LaBraM, BIOT, and CBraMod pretrained on massive unlabeled EEG corpora have pushed the field further. But labeled pathological EEG data remains scarce, most studies classify only binary disease presence rather than predicting continuous cognitive decline, and the models remain largely black boxes—unacceptable properties for a clinical diagnostic aid.

The STT-EEG framework tackles all of these limitations with four integrated components. First, the model is pretrained self-supervised on 111 healthy controls from the SRM resting-state EEG dataset (OpenNeuro ds003775), recorded on a 64-channel BioSemi ActiveTwo system at 512 Hz. The pretraining objective combines masked auto-encoding with temporal order prediction: 75% of EEG patches are masked using a hybrid strategy—60% via contiguous blocks of 4 to 12 patches and 40% randomly—and a Vision Transformer-style encoder with 12 layers, 12 attention heads, and a 768-dimensional embedding processes only the visible patches, while a lightweight four-layer decoder reconstructs the missing ones. A temporal contrastive loss, weighted at 0.1, forces the model to predict whether neighboring patch pairs are in correct or swapped temporal order. Roughly 48 hours of pretraining on an NVIDIA A100 GPU taught the network the basic statistics of healthy EEG—temporal dynamics, oscillatory structure, and channel relationships—without a single diagnostic label.

Second, the architecture introduces a novel Spatial Attention Module (SAM) alongside multi-head temporal self-attention. The temporal attention captures long-range dependencies across the 10-second epochs, which matters because pathological neural patterns appear intermittently across different segments of a resting-state recording. The SAM explicitly models cross-channel interactions: it applies temporal average pooling to aggregate global information for each channel, then computes spatial attention coefficients that quantify the relative importance of each of the 19 EEG electrodes. This design choice is deliberate. Alzheimer’s disease is a distributed network pathology, not a focal one—chronic alterations in oscillatory dynamics (slowing, reduced complexity, disrupted connectivity) pervade the entire recording—so the SAM identifies which channels are globally important while the temporal module tracks time-varying patterns. A position-wise feed-forward network with GELU activations refines features between attention blocks, and a classification head with global average pooling and a two-layer MLP produces the AD probability score.

Third, the framework demonstrates strong cross-hardware generalization. The pretraining data used a 64-channel BioSemi system, while the downstream Alzheimer’s dataset (OpenNeuro ds004504) was recorded on a 19-channel Nihon Kohden EEG-2100 clinical system at AHEPA General Hospital in Thessaloniki, Greece. The researchers harmonized the datasets by restricting recordings to the 19 electrodes common to both montages under the international 10-20 system, resampling everything to 250 Hz, segmenting into non-overlapping 10-second epochs, and applying epoch-wise z-score normalization per channel. Independent validation on a third dataset (OpenNeuro ds004148, 60 healthy participants on a 64-channel BrainProducts system) showed the pretrained representations transfer effectively, achieving 92.18% accuracy and 95.34% AUC-ROC across different hardware and task paradigms—a critical property for clinical deployment, where EEG systems differ from hospital to hospital.

The headline results are impressive. Fine-tuned with subject-wise stratified 5-fold cross-validation on 65 subjects (36 Alzheimer’s patients, 29 cognitively normal controls), STT-EEG achieved 96.42% accuracy, 98.21% AUC-ROC, 95.83% sensitivity, and 97.14% specificity, with a Matthews Correlation Coefficient of 0.929. The AD and CN groups were matched in age, sex, and education (all p-values above 0.05), so the model was not simply learning demographic confounds. In a multi-class extension distinguishing Alzheimer’s disease, frontotemporal dementia, and healthy controls, the model reached 93.18% accuracy, outperforming LaBraM by roughly 3 percentage points. It also beat 15 baseline methods spanning handcrafted-feature pipelines, supervised deep learning, and EEG foundation models: the best handcrafted approach achieved 95.86%, EEGConformer 94.22%, LaBraM 92.24%, and recent spatial-modeling architectures LUNA and Trace 91.82% and 92.34% respectively. The difference from LaBraM was statistically significant (p = 0.018 by McNemar’s test, p = 0.027 after Benjamini-Hochberg FDR correction).

Ablation studies reveal exactly where the performance comes from. Removing self-supervised pretraining costs the most—a 7.08% accuracy drop from 96.42% to 89.34%—confirming that general EEG representations learned from 111 healthy controls transfer powerfully to the scarce labeled AD task. Removing EEG-specific data augmentation costs 6.30%, and removing the SAM costs 4.86%. The augmentation pipeline is carefully engineered to preserve pathological signatures while simulating real-world variability: Gaussian noise at approximately 20 dB signal-to-noise ratio, channel dropout at 10% probability to simulate poor electrode contact, temporal warping to mimic arousal fluctuations, magnitude scaling to neutralize amplitude differences from skull thickness and impedance, and same-class Mixup to address class imbalance without creating physiologically implausible blends of pathological and normal signals. The SSL-pretrained model also converges in roughly 30 epochs versus over 60 for a supervised baseline, with a final validation loss 34% lower.

Perhaps most strikingly, the model is interpretable—and what it learned matches known neurology. Attention rollout and leave-one-channel-out perturbation analysis identified the temporal (T3-T6), parietal (P3-Pz-P4), and occipital (O1-O2) regions as the most informative electrodes, consistent with the temporoparietal neuropathology of Alzheimer’s disease. Perturbation analysis confirmed this: suppressing channel T4 caused the largest AUC drop at 8.2%, followed by Pz at 7.6% and T5 at 7.1%, while frontal channels showed minimal impact of only 1 to 2%. The convergence between intrinsic attention-based analysis and extrinsic model-agnostic perturbation provides strong validation that the network learned biologically meaningful representations rather than dataset artifacts. The framework also predicts continuous Mini-Mental State Examination (MMSE) scores from a shared representation, achieving a Pearson correlation of 0.872 with a mean absolute error of 2.34 points—within the roughly 2-point test-retest reliability of the MMSE itself, meaning the model explains valid variation in cognitive functioning within the psychometric constraints of the test.

The authors are appropriately candid about limitations. The AD dataset is small at 65 subjects, typical of clinical EEG studies, and larger multi-center cohorts will be needed before the findings are clinically reliable. The MMSE prediction error means the model may not detect subtle longitudinal cognitive changes, so it should be viewed as a supportive screening tool rather than a replacement for clinical assessment or neuroimaging. Future work will explore multimodal integration with MRI and speech biomarkers, longitudinal modeling of cognitive decline, high-density EEG, and privacy-preserving federated learning. Still, the study demonstrates something genuinely important: that self-supervised spatio-temporal representation learning, combined with interpretable attention mechanisms, can turn a cheap and ubiquitous recording technology into a credible AI-driven pathway for early Alzheimer’s detection—one that could be rapidly adopted in clinics where MRI and PET simply are not available.

Subject of Research: self-supervised spatio-temporal transformer for interpretable resting-state EEG analysis in Alzheimer's disease

Article Title: Interpretable self-supervised transformers for resting-state EEG analysis in Alzheimer's disease

Article References: Zareen, S. S., Alzaben, N., Ahmad, U., Alam, T. M., Algarni, A., & Wang, J. (2026). Interpretable self-supervised transformers for resting-state EEG analysis in Alzheimer's disease. Frontiers in Neuroinformatics, 20, Article 1899573. https://doi.org/10.3389/fninf.2026.1899573

Image Credits: AI Generated

DOI: 10.3389/fninf.2026.1899573

Keywords: Alzheimer's disease, EEG, self-supervised learning, transformer, spatio-temporal attention, MMSE prediction, interpretability, transfer learning, masked auto-encoding, deep learning, neuroinformatics, early detection

The framework was developed by Faizan Ullah and colleagues at Forschungszentrum Jülich, RWTH Aachen University, Arab Open University, and Yeungnam University. Its purpose is deliberately different from the usual goals of medical imaging AI. Rather than deciding whether a tumor is present or delineating its boundaries for clinical use, NeuroFAIR Curator acts as a data steward. It enriches metadata, screens technical annotation consistency, characterizes image quality, flags statistically unusual masks, prioritizes samples for human review, and scores the dataset against the FAIR principles of Findability, Accessibility, Interoperability, and Reusability. The authors are explicit that the system does not determine the clinical correctness of tumor annotations; it surfaces computational signals that experts can then investigate.

The test case was BRISC 2025, a public dataset of 6,000 two-dimensional contrast-enhanced T1-weighted MRI images distributed as JPEG files with binary PNG segmentation masks. The images span four categories—glioma, meningioma, pituitary tumor, and no tumor—across axial, coronal, and sagittal anatomical planes. NeuroFAIR Curator generated 6,000 sample-level curation records linked with 15,586 external manifest entries, producing 21,586 unified records, and processed 4,793 image-mask pairs. The resulting manifest serves simultaneously as a dataset description and as a curation log, carrying inferred sequence descriptors, anatomical planes, image statistics, mask statistics, consistency checks, quality flags, and model-derived signals for every sample.

Two neural networks form the computational core of the label and mask screening. A ResCNNClassifier, trained for 120 epochs with class weighting, label smoothing, MixUp augmentation, and an exponential moving average of weights, achieved 98.30 percent accuracy, 98.53 percent macro F1, and 99.79 percent macro AUC on the held-out test set. Per-class F1 ranged from 0.9736 for meningioma to a perfect 1.0000 for the no-tumor class. The classifier’s confidence and uncertainty signals, together with any disagreement between the recorded label and the model prediction, are stored as curation signals rather than diagnostic outputs. A UNet segmenter, trained for 90 epochs with a combined binary cross-entropy, Dice, and focal loss, achieved a Dice score of 82.30 percent, an IoU of 72.69 percent, precision of 83.64 percent, recall of 84.61 percent, a Hausdorff distance of 7.84 pixels, and an HD95 of 4.11 pixels at a validation-selected threshold of 0.95. Boundary quality reached 0.9964.

Explainability was built into the workflow. Grad-CAM saliency maps were generated from the final convolutional layer for the 20 most uncertain test images, and their fidelity was quantified using deletion area under the curve, or DAUC, which measures how much class confidence drops when the most salient pixels are progressively deleted. The mean DAUC was 0.2996, with per-class means of 0.3145 for glioma and 0.2933 for meningioma. Lower values indicate that the importance ordering is faithful to what the network actually relies on. The authors stress that these maps quantify faithfulness of the explanations but do not constitute clinical localization validation.

One of the most intriguing findings concerns mask morphology. The framework stratified supplied mask-area ratios by tumor class and anatomical plane and applied class- and plane-specific 1.5 times interquartile-range limits. This exploratory screening flagged 181 of 4,793 masks, or 3.78 percent, as statistically atypical. The authors are careful to note that these flags are not confirmed annotation errors; they may reflect tumor size, slice position, morphology, annotation convention, or genuine annotation problems, and without clinical-reader assessment these explanations cannot be distinguished. The flags are retained as review candidates rather than grounds for automatic exclusion, a stance consistent with prior research showing that inter-observer variability in tumor delineation can be substantial.

Technical integrity checks told a reassuring story about the dataset’s file-level organization. No missing tumor masks, empty tumor masks, dimension mismatches, filename inconsistencies, unknown planes, or corrupted samples were detected. Image-quality screening, by contrast, revealed a small but real tail of problematic images: 51 low-contrast cases, 60 blur outliers, 208 intensity outliers, and 180 anomalies detected by an Isolation Forest. Across the corpus, 456 samples carried at least one flag and 184 carried multiple flags simultaneously, while 5,544 samples had zero flags.

The framework’s most distinctive contribution is the Curation Uncertainty Index, or CUI, a transparent seven-component score that fuses classification uncertainty, segmentation uncertainty, mask abnormality, metadata incompleteness, image-quality anomaly, image-mask mismatch, and metadata-validation error. The baseline weights were prespecified as a hierarchical prior rather than fitted to test data, with model uncertainty receiving 0.36, statistical and structural mask evidence 0.16, reuse-quality evidence 0.28, and sparse technical validation signals 0.20. A one-at-a-time sensitivity analysis perturbed each weight by up to 50 percent and showed that the ranking remained stable under most perturbations, with Spearman correlations above 0.99. The largest ranking change occurred when segmentation uncertainty was reduced by half, highlighting which signals drive the prioritization.

Active curation prioritization demonstrated the practical value of the CUI. At a review budget of just 1 percent of the dataset, the CUI ranking captured 49 of the 456 uniquely flagged samples, a 10.75 percent capture rate that represents a 99 percent reduction in review workload compared with random inspection. At budgets of 5 and 10 percent, capture rose to 15.13 and 17.11 percent respectively. The CUI assigned 5,001 samples, or 83.35 percent, to low curation priority and 999 samples, or 16.65 percent, to moderate priority, giving curators a concrete, defensible ordering for their limited expert time.

The FAIR readiness assessment, scored on an 18-sub-criterion rubric, showed the largest gains where the dataset was weakest. Findability improved from 87.50 to 100.00 and Interoperability from 36.00 to 91.98, reflecting the addition of structured descriptors, harmonized metadata, quality indicators, and provenance fields. Accessibility and Reusability remained unchanged at 95.00 and 95.98 under the declared rubric, a pattern the authors attribute to the specific instrument and dataset rather than a universal result. Taken together, NeuroFAIR Curator converts a widely reused public brain tumor MRI corpus into a metadata-enriched, technically checked, quality-screened, review-prioritized, and FAIR-aligned neuroinformatics resource. The authors note that future work should evaluate the framework with blinded expert review, multicenter DICOM or NIfTI datasets, three-dimensional annotations, and external calibration across different scanners and acquisition protocols, since the operating threshold and CUI weights are specific to the BRISC 2025 validation distribution rather than universal parameters.

Subject of Research: AI-assisted metadata enrichment, quality screening, curation prioritization, and FAIR readiness assessment of public brain tumor MRI data

Article Title: NeuroFAIR curator: an AI-assisted framework for metadata enrichment, quality screening, curation prioritization, and FAIR readiness of brain tumor MRI data

Article References: Ullah, F., Abbas, Z., Abrar, M., Amin, F., & Shah, N. J. (2026). NeuroFAIR curator: an AI-assisted framework for metadata enrichment, quality screening, curation prioritization, and FAIR readiness of brain tumor MRI data. Frontiers in Neuroinformatics, 20, Article 1907796. https://doi.org/10.3389/fninf.2026.1907796

Image Credits: AI Generated

DOI: 10.3389/fninf.2026.1907796

Keywords: brain tumor MRI, FAIR data, metadata enrichment, data curation, neuroinformatics, annotation integrity, uncertainty quantification, active learning, image quality control, BRISC 2025, Curation Uncertainty Index, deep learning

The problem stems from a fundamental property of EEG. Electrodes do not measure absolute electrical potentials; they measure potential differences relative to a reference. Because no perfect, silent reference exists, researchers must choose one, and the choice shapes everything downstream. Previous work by the same group, based at the Universidad Autónoma del Estado de Morelos and the Universidad Nacional Autónoma de México, had already established that the median reference, in which each electrode’s signal is expressed relative to the median of all electrodes at every instant, produces functional brain networks that closely resemble those obtained from magnetoencephalography, a technique that is essentially reference-free. The median reference, it seemed, was the safest choice.

But EEG data also need to be filtered, typically into the classical frequency bands such as delta, theta, alpha, beta, and gamma, which are associated with different cognitive and physiological states. That raises a question that sounds almost trivial: should the signals be filtered first and then re-referenced to the median, a scheme the authors call FR, or re-referenced first and then filtered, a scheme they call RF? Intuition might favor RF, since re-referencing only once reduces the computational workload. The new study demonstrates that this shortcut comes at a steep and sometimes irreparable price.

The team, led by J. Daniel Arzate-Mena, Paola V. Olguín-Rodríguez, and Markus F. Müller, tested both schemes on three kinds of data: whole-night sleep EEG recordings from ten healthy young men, synthetic signals from controlled numerical models, and surrogate data that mimic the statistical properties of real EEG while containing no genuine correlations at all. The numerical models included Nf-Tori, which are linear superpositions of two thousand harmonic oscillations whose power spectra could be tuned to follow power laws of varying steepness, and IAAFT surrogates, which preserve the amplitude distribution and power spectrum of real recordings while destroying any true interdependence between channels. This combination allowed the researchers to know exactly what the correct answer should be, and to see precisely how each preprocessing scheme deviated from it.

The results were striking. When re-referencing was performed before filtering, the RF scheme, spurious positive cross-correlations appeared throughout the network, and they grew systematically stronger in the faster frequency bands. In the gamma band, during deep sleep, the average displacement of correlation values reached about 0.5, with individual estimates climbing as high as 0.75. The true spatial correlation structure was overshadowed by this flood of artificial correlations, simulating a kind of hypersynchronized brain dynamics that simply does not exist. In contrast, the FR scheme introduced only small, systematic negative shifts of roughly minus 0.025, which were not statistically significant.

Most alarming was the finding that the RF-induced artifacts are irreversible. The researchers applied a sophisticated statistical correction known as the Significant Average Correlation matrix, or SAC-matrix, which compares each pair of channels against surrogate data at a Bonferroni-corrected significance level to separate genuine correlations from random ones. For the FR scheme, this correction worked beautifully, essentially eliminating the tiny residual distortions. But for the RF scheme, even the SAC-matrix could not repair the damage. The artificial correlations survived the correction, meaning they become indistinguishable from genuine brain connectivity. Once the redundant information has been embedded into every channel by the flawed pipeline, no post-hoc statistical trick can tease it apart.

The mechanism behind the effect lies in the peculiar behavior of the median reference itself. At any given moment, the median of an odd number of channels equals the potential of one specific electrode, which is then subtracted from all the others. The researchers measured how long each electrode serves as the reference before another takes over. For steep power spectra, they found that high-amplitude, slow-frequency components can dominate the median signal for hundreds of consecutive sampling points. During these extended intervals, the same slow waveform is effectively stamped into every channel, embedding redundant information across the entire recording. When the data are subsequently filtered into fast frequency bands, the mismatch between the slow time scale of the reference signal and the fast time scale of the band creates the spurious correlations, with the effect most pronounced in the beta and gamma ranges.

The severity of the problem is governed by the power spectral exponent beta, the slope of the signal’s power spectrum. When the team tuned their synthetic signals to follow power laws from white noise, with beta equal to zero, up to steep red noise with beta equal to three, the artificial correlations grew with every increase in steepness. This matters because beta values of two or greater are not exotic. They characterize slow-wave sleep, the post-ictal period following focal epileptic seizures, anesthesia, neonatal recordings, and patients in vegetative states or coma. In the sleep data, the spectral slope correlated strongly with the amount of spurious gamma-band correlation under the RF scheme, with a correlation coefficient of roughly 0.85, while the FR scheme showed no such dependence. Deep sleep stages, with slopes approaching minus three, were hit hardest.

The implications reach well beyond sleep research. Any study that constructs functional brain networks from multivariate EEG recordings, whether investigating epilepsy, disorders of consciousness, infant development, or insomnia, could be producing distorted connectivity maps if the preprocessing order is wrong. The authors also showed that the global average reference, an alternative proposed back in 1950, suffers from its own flaw: high-amplitude channels dominate the average and induce spurious coupling in low-amplitude channels, a pattern that even the SAC-matrix cannot fully correct. The median reference, by contrast, is far more robust to amplitude differences caused by muscle or eye artifacts.

The team’s recommendation is clear: filter first, then re-reference, and always apply surrogate-corrected cross-correlations when analyzing signals with steep power spectra. The FR scheme demands more computation, since each filtered signal must be re-referenced individually, but the resulting distortions are an order of magnitude smaller and can be almost completely eliminated. Although the study was conducted on EEG, the authors emphasize that the principle applies to any multivariate recording in which spatial re-referencing and band-pass filtering are applied sequentially. In an era when brain connectivity studies inform everything from consciousness research to clinical diagnosis, the message is a sobering reminder that in signal processing, as in life, the order of operations is not a technical detail. It can be the difference between discovering how the brain truly communicates and chasing ghosts that the pipeline itself created.

Subject of Research: The effect of preprocessing step order on spurious correlations in EEG functional network analysis

Article Title: Irreversible functional network deformation induced by preprocessing scheme order in multivariate EEG recordings

Article References: Arzate-Mena, J. D., Olguín-Rodríguez, P. V., Corsi-Cabrera, M., López-Alvarado, E. A., & Müller, M. F. (2026). Irreversible functional network deformation induced by preprocessing scheme order in multivariate EEG recordings. Frontiers in Neuroinformatics, 20, Article 1907920. https://doi.org/10.3389/fninf.2026.1907920

Image Credits: AI Generated

DOI: 10.3389/fninf.2026.1907920

Keywords: EEG, preprocessing, functional connectivity, median reference, spurious correlations, spectral slope, sleep, surrogate data, cross-correlation, neuroinformatics, band-pass filtering, brain networks

The research, led by Valentin Fernandez and colleagues at INSERM U1084 and the University of Poitiers, together with collaborators in Coimbra and Samarkand, addresses a subtle but consequential blind spot. Current evaluation of pose estimation pipelines relies almost exclusively on frame-wise spatial accuracy metrics such as root mean square error (RMSE), percentage of correct keypoints (PCK) and mean average precision (mAP). These metrics measure how closely the AI’s predicted landmark positions match human annotations in individual frames. What they fail to capture, the authors contend, is whether the resulting trajectories are reliable enough to yield trustworthy kinematic biomarkers downstream.

The problem becomes acute when those biomarkers involve derivatives. Quantities such as velocity, acceleration and jerk, the rate of change of acceleration, are computed by temporal differentiation of the reconstructed trajectories. Differentiation acts as a noise amplifier: high-frequency tracking jitter and frame-to-frame fluctuations that barely register in a spatial accuracy score can be magnified into substantial distortions of the derived movement features. Two pipelines with nearly identical localisation accuracy may therefore produce trajectories of very different quality once temporal consistency, occlusion handling and kinematic fidelity are taken into account.

To close this gap, the team developed a reliability-centred evaluation framework built around a new composite metric, the Motor Biomarker Reliability Composite Score, or MBRCS. The score integrates four complementary dimensions of trajectory reliability: spatial accuracy, temporal stability, occlusion robustness and kinematic fidelity. Each component is normalised to a scale from zero to one using robust percentile-based scaling, in which values are clipped to the fifth and ninety-fifth percentiles of the evaluated distribution to reduce sensitivity to outliers. Higher values indicate greater trajectory reliability. The authors emphasise that a normalised score near zero reflects relative positioning within the benchmark, not an absence of capability.

The weighting scheme assigns greater importance to temporal stability and kinematic fidelity, a choice made a priori because derivative-based kinematic descriptors are particularly vulnerable to temporal fluctuations and noise amplification. Sensitivity analyses using alternative weighting configurations, including uniform weighting and accuracy-prioritised schemes, produced highly consistent architecture rankings with Spearman correlation coefficients above 0.9, indicating that the composite score is not strongly dependent on the precise weights chosen. The MBRCS is intended as a comparative benchmarking tool for selecting pipelines, not as an absolute measure of performance, and computational efficiency was deliberately excluded from the score.

The framework was tested on a demanding experimental paradigm: the Montoya staircase task, in which mice retrieve food pellets from descending staircases on either side of a central platform. Adult C57BL/6 mice, twelve control animals and twelve with unilateral lesions of the primary motor cortex, were recorded from a lateral view at 120 frames per second with a resolution of 1,280 by 1,024 pixels. Around 4,800 frames from 24 behavioural recordings were manually annotated, covering landmarks including the shoulder, elbow, wrist, forearm, base of the fingers and fingertips. The task is a particularly rigorous benchmark because skilled reaching involves small-amplitude distal limb displacements, rapid temporal transitions and frequent partial occlusions caused by self-occlusion and contact with the apparatus.

Nine DeepLabCut-based convolutional neural network architectures were trained and tested under identical conditions using DeepLabCut version 3.0 with a PyTorch backend, spanning ResNet-50, ResNet-101 and ResNet-152, MobileNetV2, EfficientNet-B0 and EfficientNet-B3, and several DLCRNet-based configurations. All models were trained for 600,000 iterations on the same annotated dataset with the same split, optimisation pipeline and post-processing, so that performance differences could be attributed to architecture rather than training strategy. Predicted landmarks with a probability below 0.9 were treated as low-confidence detections; short gaps of up to five consecutive frames were reconstructed by linear interpolation, while longer interruptions, which accounted for less than 3 percent of low-confidence segments, were excluded from kinematic analysis. Trajectories were then smoothed with a Gaussian filter of sigma equal to 2 frames.

The results revealed a striking dissociation. Spatial localisation metrics remained comparatively clustered across architectures, with RMSE ranging only from 2.80 to 3.58 pixels, PCK at a 5-pixel threshold from 90.70 to 95.90 percent, and mAP from 85.63 to 91.07 percent. Yet the reliability dimensions told a very different story. Temporal stability varied substantially, and normalised jerk values, the downstream biomarker used to assess kinematic fidelity, ranged from 1.60 for the best-performing DLCRNet_ms5 to 3.85 for MobileNetV2. When all four dimensions were combined, MBRCS values spanned almost the entire normalised range, from 0.00 to 0.99, with DLCRNet_ms5 at the top and MobileNetV2 at the bottom.

Crucially, these reliability differences had functional consequences. In an external validation step deliberately excluded from the MBRCS itself, the researchers assessed whether trajectories with higher intrinsic reliability better preserved biologically meaningful differences between the control and lesion groups. Using Cohen’s d effect sizes as the primary measure of known-group validity, complemented by receiver operating characteristic analysis with logistic regression classifiers under stratified five-fold cross-validation, they found that architectures with higher temporal stability and kinematic fidelity showed greater separation of jerk distributions between groups. DLCRNet_ms5 and DLCRNet_stride32 achieved both larger effect sizes and higher AUC values, whereas MobileNetV2 showed the weakest discriminative performance, with an AUC of approximately 0.60.

The authors are careful to frame their contribution as complementary rather than substitutive. Spatial localisation accuracy remains meaningful, and the best-performing architectures in the composite score were also the most accurate localisers, but the additional dimensions exposed differences that frame-wise metrics alone could not reveal. The framework’s limitations are acknowledged: the evaluation was restricted to DeepLabCut-based architectures and a single downstream biomarker, robustness was assessed primarily under occlusion-related perturbations rather than motion blur or reduced frame rate, and variability from independent training repetitions was not evaluated. Nevertheless, the researchers argue that the methodology offers a structured, reproducible path toward benchmarking pose estimation pipelines by the quality of the kinematic information they preserve, a shift with particular relevance for studies of subtle motor impairment, longitudinal monitoring and acquisition conditions plagued by partial occlusions, where the reliability of the trajectory, not just the accuracy of the pose, determines whether a biomarker can be trusted.

Subject of Research: A reliability-centred evaluation framework for markerless pose estimation in mouse movement analysis

Article Title: From pose estimation accuracy to motor biomarker reliability: a reliability-centred evaluation framework for markerless movement analysis

Article References: Fernandez, V., Abdiyeva, K., Brot, S., Perreira de Almeida, L., Gaillard, A., & Fernandez-Maloigne, C. (2026). From pose estimation accuracy to motor biomarker reliability: a reliability-centred evaluation framework for markerless movement analysis. Frontiers in Neuroinformatics, 20, Article 1790241. https://doi.org/10.3389/fninf.2026.1790241

Image Credits: AI Generated

DOI: 10.3389/fninf.2026.1790241

Keywords: markerless pose estimation, DeepLabCut, motor biomarkers, kinematic analysis, trajectory reliability, Montoya staircase task, Normalized Jerk Score, occlusion robustness, temporal stability, neuroinformatics, movement smoothness, convolutional neural networks

The tool, called BVLab-Analyzer, is described in a study published in Frontiers in Neuroinformatics by Ying Hu of Hubei University of Education, Xiaojun Wang of Hainan University, and colleagues at the Britton Chance Center for Biomedical Photonics at Huazhong University of Science and Technology. It is a standalone, zero-code graphical application for Windows that carries a terabyte-scale brain dataset through every stage of analysis: importing data, annotating training examples, training a deep neural network, running whole-brain inference, stitching results together, and exporting standardized morphological reconstructions and statistics. Crucially, the entire workflow runs on a single consumer-grade graphics card.

That last point is more radical than it may sound. Most existing deep-learning frameworks for whole-brain analysis—Cellfinder, nnU-Net-based pipelines, and others—assume access to multi-GPU clusters or large-memory enterprise servers. For a typical biology laboratory equipped with one workstation and a gaming-class GPU, whole-brain quantification has effectively been out of reach. BVLab-Analyzer tackles the hardware problem with an adaptive 3D block-wise tiling scheme: the software automatically slices a complete brain volume, which can span three to thirteen terabytes, into tens of thousands of independent sub-volume blocks sized to fit within 8 to 14 gigabytes of GPU memory. Cell detection uses anisotropic blocks of 272 by 272 by 144 voxels to capture the wide footprint of neuronal cell bodies, while neuron and vessel tracing use isotropic 192-cubed blocks to preserve continuity along fibers and vessels in depth.

The architecture itself is a hybrid of C++ for compute-intensive operations and Python for the deep-learning logic, wrapped in a Qt graphical interface and compiled into a pre-bundled executable that eliminates the notorious pain of configuring Python environments and CUDA versions. Data flows through six tabbed modules—Data I/O, Annotation, Training, Inference, Post-processing, and Statistics—all coordinated by a central project file that preserves parameters and fault-tolerant state. The software reads multi-page TIFF stacks, the BioimageVision format, and the cloud-optimized OME-Zarr format, automatically parsing multi-resolution pyramids so users can move between overview and detail levels during visualization, which is rendered in real time through the VTK library.

Training a deep network normally demands enormous annotated datasets, and here the researchers introduced what may be the software’s most elegant trick. A built-in K-Means clustering algorithm extracts grayscale histogram and maximum-intensity-projection features from every sub-block in a whole brain and groups them into one to two thousand clusters. From each cluster, the block nearest the center is offered as a candidate, and the researcher hand-picks only about one to two hundred representative blocks for sparse annotation. This data-driven sampling reduced the required annotation volume by roughly an order of magnitude compared with experience-based random selection. In the reported experiments, training sets of just 166 blocks for cells, 86 for neurons, and 120 for vessels were enough.

Annotation itself is accelerated by what the authors call Smart Annotation tools. For point-like cell bodies, a single click at the soma center registers a 3D coordinate in under a second and stores it as a root node in the standard SWC topology format. For linear structures such as axons, dendrites, and blood vessels, users place sparse anchor points along the fiber, and the software auto-interpolates the intermediate skeleton path with parent-child links compliant with the SWC specification. Large vessels that appear hollow or faint after tissue clearing get a dedicated ray-casting mode that projects orthogonal rays from reference points, detects the vessel wall boundary by local intensity drops, and computes the true cross-sectional midpoint and radius. Sparse cell points are then converted to Gaussian blobs and skeleton trajectories into tubular Gaussian volumes, producing dense training masks without ever requiring pixel-level contour drawing.

For inference, the trained 3D U-Net sweeps the whole brain with overlapping sliding windows whose overlaps are tuned to biology: 32 by 32 by 20 voxels for cells, roughly one and a half to two soma diameters, and 4 by 4 by 2 for the thin linear structures. After prediction, the topologies of all blocks are merged back into whole-brain coordinate space. Perhaps the most practically beloved feature will be the interrupt-and-resume engine. Whole-brain prediction can take up to 129 hours, and on shared lab workstations interruptions are inevitable. The software logs completed sub-volumes in real time, so after a power outage or a colleague borrowing the machine for the day, it simply skips the finished blocks and continues—validated in the study by pausing a neuron reconstruction after 48 hours, using the workstation for office work, and resuming without any loss of progress.

The performance numbers are competitive with the specialized state of the art. On a 3.6-terabyte, 16-bit c-Fos cell-detection volume, the pipeline achieved a median precision of 0.863 and an F1-score of 0.868, despite grayscale means that stepped from several hundred to nearly 3,000 across the brain. On a 12.36-terabyte GFP-labeled neuron dataset—the largest tested—the median recall reached 0.893 with an F1 of 0.884, maintaining continuous tracing even under ultra-low-contrast fiber signals. On a 0.47-terabyte vascular volume, the median recall hit 0.944 with an F1 of 0.899, though large-vessel luminal cavities remain a known weak spot. Total training time per task was kept between 1.9 and 2.8 hours on a single RTX 4090, with GPU memory stable at 8 to 14 gigabytes throughout.

What truly distinguishes BVLab-Analyzer, the authors argue, is scope. Cell somata, neuronal processes, and the microvasculature are not isolated objects but together form the neurovascular unit, the functional target of studies of neurovascular coupling. Until now, a laboratory wanting to correlate all three had to juggle separate specialized tools—ClearMap, Cellfinder, TrailMap, Vaa3D—with incompatible output formats and manual bridging between pipelines. BVLab-Analyzer unifies all three target structures in one zero-code environment and exports everything in standard SWC format, readable directly in Vaa3D and neuTube, with statistics such as cell density, total fiber length, branch-point counts, bifurcation angles, and vessel diameter distributions exported as CSV tables for immediate plotting.

The team is candid about limitations. The software processes each fluorescence channel with a dedicated model and cannot separate mixed signals within a single channel; it currently handles only static, ex vivo rodent whole-brain data, runs only on Windows, and cross-region and cross-specimen generalization has not yet been formally evaluated. Difficult regions—dense cell clusters, extremely low signal-to-noise fibers, hollow large vessels—still rely on built-in interactive proofreading tools rather than automatic quality detection, though uncertainty estimation and adaptive post-processing are planned. A Python API and cross-platform support are on the roadmap. Even so, the package marks a genuine democratization of whole-brain analysis: the kind of terabyte-scale quantification that once demanded a computing cluster and a programmer now fits inside a single desktop window, a single GPU, and a workflow that a bench biologist can drive entirely by hand.

Subject of Research: A unified zero-code software toolbox for terabyte-scale whole-brain cell detection, neuron reconstruction, and vascular network extraction

Article Title: BVLab-analyzer: a unified visual toolbox for terabyte-scale whole-brain cell detection, neuron reconstruction, and vascular network extraction

Article References: Hu, Y., Wang, X., Sun, P., Shen, Y., & Quan, T. (2026). BVLab-analyzer: a unified visual toolbox for terabyte-scale whole-brain cell detection, neuron reconstruction, and vascular network extraction. Frontiers in Neuroinformatics, 20, Article 1925112. https://doi.org/10.3389/fninf.2026.1925112

Image Credits: AI Generated

DOI: 10.3389/fninf.2026.1925112

Keywords: whole-brain imaging, deep learning, BVLab-Analyzer, 3D U-Net, neuron reconstruction, vascular segmentation, cell detection, light-sheet microscopy, SWC format, neuroinformatics, GPU computing, zero-code software

The core insight behind the work is deceptively simple. Electroencephalography, the scalp electrodes that record the brain’s electrical rhythms, is notoriously fragile outside the laboratory. Electrodes shift, muscles fire, power lines inject 50 or 60 hertz hum, and contact impedance drifts over hours of monitoring. Any of these can silently corrupt the incoming signal, producing a prediction that looks confident but rests on garbage. Existing systems either clean artifacts in a separate preprocessing step or ignore signal quality entirely, and the study’s authors argue that this disconnect between benchmark performance and clinical reliability has held the field back.

CLSP-REQA, short for Closed-Loop Seizure Prediction with Real-time EEG Quality Assessment, embeds a lightweight one-dimensional convolutional network directly inside the prediction pipeline. This module, dubbed REQA, runs in parallel with the main prediction backbone and emits a scalar quality score between 0 and 1 for every 10-second window of EEG data. Crucially, the score is not hand-crafted. It is trained end-to-end as part of the full prediction objective, receiving gradient signal only indirectly through its role in the downstream confidence fusion, which means the system needs no labeled artifact annotations—annotations that are rarely available in clinical EEG archives.

The prediction backbone itself combines two established architectures. Twelve cascaded bidirectional Mamba blocks, drawn from the EEGMamba encoder, model long-range temporal dependencies with linear computational complexity, avoiding the quadratic cost that limits transformers on long recordings. A two-layer bidirectional LSTM then refines these features, capturing fine-grained local temporal transitions. The authors are candid that this combination is a competent engineering choice rather than a new architectural primitive; the novelty they claim lies in coupling a learned, end-to-end quality estimator to a tiered, safety-bounded confidence-fusion rule designed specifically to gate a closed-loop intervention decision.

That fusion rule, called ECLO for Evidence-Calibrated Layered Output, works in three tiers. When signal quality exceeds 0.8, the backbone’s seizure probability is trusted directly. At moderate quality between 0.5 and 0.8, a super-linear penalty makes the system increasingly conservative as quality degrades. Below 0.5, confidence is hard-capped at 0.3 regardless of the raw prediction, preventing artifact-driven false alarms and raising a quality warning flag. The authors contrast this with simple linear weighting, which treats quality degradation symmetrically and imposes no safety ceiling—a critical gap for closed-loop neurostimulation, where a spuriously high seizure probability could trigger unnecessary stimulation with real clinical risks.

Because the quality score receives no direct supervision, the team designed three experiments to test whether it genuinely tracks signal quality rather than collapsing into a redundant proxy for the prediction itself. In the first, they synthetically corrupted clean test windows with six artifact types—Gaussian noise, muscle-like bursts, electrode-displacement jumps, line noise, baseline drift, and signal clipping—at graded intensities. The learned score declined monotonically with corruption severity across all six types, with Spearman correlations ranging from 0.671 for line noise to 0.802 for Gaussian noise. In the second, the score correlated strongly with an independent, model-free signal quality index computed from variance anomalies and frequency-band power ratios. In the third, the score showed essentially no correlation with either the seizure probability or the ground-truth label, confirming it had not become a second prediction channel in disguise.

Evaluation was deliberately harsh. The team used strict cross-patient validation on the CHB-MIT Scalp EEG Database, 940 hours of recordings from 23 pediatric patients containing 198 seizures, and on the SIENA Scalp EEG Database, 128 hours from 14 adult patients with 47 seizures. Training and test sets never shared a patient, and results were averaged over 25 independent runs across five folds and five seeds. On CHB-MIT, CLSP-REQA achieved an AUC-ROC of 0.7426, exceeding the unadapted cross-patient baseline of 0.69 reported by prior work, while using only 16 EEG channels compared with 23 in the baseline system. On SIENA, it reached an AUC of 0.7012, surpassing the best domain-adapted cross-patient result of 0.61 on the same dataset—despite the fact that domain-adaptation methods are given access to unlabelled target-patient data that CLSP-REQA never sees.

The authors are careful about what that comparison means. They frame quality-aware confidence gating as a complementary, deployment-friendly lever for cross-domain robustness rather than a substitute for domain adaptation, noting that a domain-adapted variant of their own system might perform better still. They also flag statistical limitations: the 25 runs are not fully independent because seed-repeats within a fold share the same test patients, so they additionally report a conservative fold-level analysis in which all five fold means exceed the 0.69 baseline, yielding the most extreme possible one-sided Wilcoxon p-value at that sample size.

Speed matters for a closed-loop system, and the benchmarks are striking. The full pipeline processes a single 10-second window in 7.48 milliseconds on an NVIDIA A100 GPU, a real-time margin of roughly 1,337-fold, and in 47.71 milliseconds on a representative low-power edge platform—still about 210 times faster than the data-acquisition window. The entire model contains just 3.5 million parameters. Every inference step emits a structured four-tuple: the seizure probability, the signal quality score, the fused confidence, and a SHAP attribution vector identifying which EEG channels drove the prediction. Based on confidence, the system recommends one of three intervention levels—immediate, alert, or monitor—formatted for direct interfacing with closed-loop neurostimulator protocols such as the implanted devices already used in responsive neurostimulation therapy.

The team is equally explicit about what the study does not establish. Both datasets are retrospective recordings from controlled hospital settings, so the work is presented as a feasibility demonstration rather than evidence of prospective, real-world clinical readiness. The ECLO thresholds were set empirically rather than derived from a theoretical calibration objective, though a sensitivity analysis showed the chosen operating point sits on a broad performance plateau, and a head-to-head comparison against temperature scaling and isotonic regression confirmed that only ECLO structurally bounds its confidence on low-quality windows. Prospective, multi-center validation under ambulatory conditions, where artifacts and electrode aging are far more heterogeneous, remains the necessary next step before quality-aware seizure prediction can reach the clinic.

Subject of Research: A quality-aware closed-loop deep learning framework for real-time epileptic seizure prediction from scalp EEG

Article Title: CLSP-REQA: a real-time quality-aware closed-loop seizure prediction framework with Mamba-BiLSTM and confidence-gated intervention

Article References: Chen, M., Wu, Q., Huang, B., Lai, X., Chen, Z., Ouyang, X., & Ren, Q. (2026). CLSP-REQA: a real-time quality-aware closed-loop seizure prediction framework with Mamba-BiLSTM and confidence-gated intervention. Frontiers in Neuroinformatics, 20, Article 1888743. https://doi.org/10.3389/fninf.2026.1888743

Image Credits: AI Generated

DOI: 10.3389/fninf.2026.1888743

Keywords: seizure prediction, epilepsy, EEG, closed-loop neurostimulation, Mamba, BiLSTM, signal quality assessment, cross-patient generalization, confidence calibration, SHAP interpretability, CHB-MIT, SIENA

The enzymes in question are the Rho-associated coiled-coil kinases, ROCK1 and ROCK2, which have emerged in recent years as critical upstream regulators of mitochondrial behavior in neurons. ROCK1 promotes Drp1-mediated mitochondrial fission, the process by which mitochondria split into daughter organelles. While some fission is essential for cellular housekeeping, excessive fission fragments the mitochondrial network and contributes to neuronal death, a mechanism documented in Parkinson’s disease where ROCK1 activation drives dopaminergic cell apoptosis. ROCK2, by contrast, suppresses the PINK1-Parkin pathway, the cellular machinery responsible for identifying damaged mitochondria and tagging them for destruction through mitophagy. When ROCK2 is overactive, worn-out mitochondria linger in the cell, spewing reactive oxygen species and draining ATP reserves.

Overexpression of both ROCK isoforms, the researchers note, leads to increased ROS production and reduced ATP generation, a double blow that compromises neuronal health and fuels the neuroinflammation characteristic of Alzheimer’s, Parkinson’s, and related disorders. Synthetic ROCK inhibitors such as fasudil have shown neuroprotective potential in animal models, reducing amyloid-beta production and improving proteomic profiles in Alzheimer’s mice. Yet their translation into long-term therapy for chronic neurological disease has been hampered by toxicity concerns, poor pharmacokinetics, and suboptimal penetration of the blood-brain barrier, the selective border that shields the brain from most circulating molecules.

Natural compounds offer a compelling alternative. Plant-derived phytochemicals are often cheaper to produce, generally better tolerated, and in some cases possess scaffolds optimized by evolution for biological activity. To explore this possibility systematically, Shandilya and Mani turned to the pharmacopeia of Indian traditional medicine. Drawing on documented neuroprotective properties and underexplored mechanisms of action, they selected eleven medicinal plants, including species such as Convolvulus pluricaulis, Centella asiatica, Alpinia galanga, and Evolvulus alsinoides, all of which have been studied for cognitive and neuroprotective effects in preclinical work.

From these plants, the team assembled a virtual library of 1,223 natural product compounds, which was then curated down to 428 unique phytocompounds for screening. A known ROCK inhibitor was included as a reference benchmark. The screening pipeline followed the standard architecture of modern computational drug discovery: molecular docking to predict how each compound fits into the ATP-binding pockets of ROCK1 and ROCK2, followed by successive layers of increasingly rigorous validation designed to weed out false positives.

Several natural phytocompounds achieved binding affinities comparable to the reference inhibitor, demonstrating that plant chemistry can compete with synthetic medicinal chemistry at these targets. The shortlist was then subjected to pharmacokinetic profiling and toxicity classification to assess drug-likeness, absorption, and safety liabilities. MM-GBSA calculations provided estimates of free binding energy in a more realistic solvent environment than docking alone, and finally, molecular dynamics simulations running for 200 nanoseconds tested whether the predicted complexes remained stable when the proteins were allowed to flex and move in water, as they do in living cells.

The results converged on a handful of standout molecules. Scopolin and lycorine emerged as promising ROCK1 inhibitors, the isoform responsible for driving pathological mitochondrial fission. Cytisoside and N-methyltyramine demonstrated strong inhibition of ROCK2, the suppressor of PINK1-Parkin mitophagy. Each of these compounds could, in principle, address one half of the mitochondrial quality-control problem: either curbing excessive fission or releasing the brake on damaged-mitochondria removal.

The most intriguing finding, however, was coumarin. This simple benzopyrone, found in many plants including several of the species screened, exhibited stable and persistent interactions with both ROCK1 and ROCK2 across the full simulation window. It combined good binding energies with high predicted blood-brain barrier permeability, addressing one of the major failings of existing synthetic inhibitors, and maintained strong dynamic stability in the 200-nanosecond trajectories, suggesting the interactions are not artifacts of a static protein structure. A dual inhibitor is particularly attractive because mitochondrial fission and mitophagy are mechanistically linked: fission facilitates the selective autophagy of damaged organelles, so a single molecule that rebalances both arms of the pathway could restore mitochondrial quality control more coherently than two separate drugs.

The authors are careful to frame these findings as a starting point rather than a therapeutic endpoint. In-silico screening, however sophisticated, generates hypotheses that must survive contact with biology. The study concludes that coumarin’s potential to restore the balance between mitochondrial fission and mitophagy, and thereby mitigate neurodegenerative disease, should be validated using in vitro assays, where the compounds can be tested against living neurons and measured for their actual effects on Drp1 recruitment, PINK1-Parkin signaling, ROS levels, and ATP production. Pharmacokinetic predictions also require laboratory confirmation, since computational models of blood-brain barrier penetration are approximations.

Even so, the work illustrates how neuroinformatics is reshaping the search for neuroprotective drugs. Rather than screening millions of synthetic molecules at enormous cost, researchers can now mine the accumulated ethnopharmacological knowledge of traditional medicine systems, filter it through structure-based computation, and arrive at a small set of testable candidates within months. If coumarin and its fellow phytocompounds perform well in cellular and animal models, they could open a path toward safer, cheaper, and more brain-accessible therapies for diseases that currently have few options capable of slowing their progression. For now, the mitochondria of neurons under threat have gained a new set of molecular allies, at least on the computer screen.

Subject of Research: In-silico screening of natural phytocompounds as ROCK1 and ROCK2 inhibitors to modulate mitochondrial fission and mitophagy in neurological diseases

Article Title: Targeting ROCK Isoforms by Natural Phytocompounds: An In-Silico Approach to Modulate Mitochondrial Fission and Mitophagy in Neurological Diseases

Article References: Shandilya, C., & Mani, S. (2026). Targeting ROCK Isoforms by Natural Phytocompounds: An In-Silico Approach to Modulate Mitochondrial Fission and Mitophagy in Neurological Diseases. Neuroinformatics, 24(4), Article 67. https://doi.org/10.1007/s12021-026-09822-0

Image Credits: AI Generated

DOI: 10.1007/s12021-026-09822-0

Keywords: ROCK1, ROCK2, mitochondrial fission, mitophagy, phytocompounds, coumarin, neurodegenerative diseases, molecular docking, molecular dynamics, Indian medicinal plants, blood-brain barrier, neuroinformatics