Machine learning models (MLMs) have been used to classify neuron subtypes based upon electrophysiological attributes. What has been less explored is using MLMs to assess differences within a neuron type, especially in the context of subtle changes induced by neuromodulation such as the rodent estrous cycle. Previous research found that the estrous cycle shifts rat nucleus accumbens (NAc) medium spiny neuron (MSN) electrophysiology, including action potential (AP), passive, and miniature excitatory post-synaptic current (mEPSC) properties. This plasticity provides a model system to investigate this question. We hypothesized that MLM classification accuracy would differ when trained across these data types, reflecting the information each encodes regarding estrous phase identity. To test this hypothesis, we extracted electrophysiological features across four estrous cycle phases from a publicly available dataset, and employed this data to train MLMs to classify estrous cycle phase origin. We found: MLMs identified estrous phase origin with up to 94
Neural data collected using brain-computer interfaces, neural implants, and emotion detection systems is analyzed by AI classifiers and agentic architectures to serve purposes such as authentication, access control, and behavioral inference, however, there exists no comprehensive, binding cybersecurity or data protection regime to regulate such neural data. The regulations that currently exist i.e., GDPR, HIPAA, the Budapest Convention, the 2025 UNESCO Recommendation on Neurotechnology Ethics, and a small number of state laws (e.g., Colorado 2024, California SB 1223, Montana, Connecticut) create a fragmented and incomplete emerging framework rather than no framework at all. In this paper, the author propose Cognitive Sovereignty architecture, an approach of governance through the combination of a legally recognized definition and technical parameters defining neural data as a new class of data which necessitates specific regulatory, adversarially sound processing frameworks, and jurisdictionally agnostic enforcement mechanisms. By conducting comparative law research, threat modeling based on STRIDE model and governance modeling, this paper highlights structural issues with the existing regimes and suggests a framework composed of Declaration on Cognitive Sovereignty, neuro-cybercrime protocol of the Budapest Convention, and AI layer-specific compliance requirements based on NIST AI RMF and the EU AI Act.
Visual inspection as the golden standard for sleep spindle detection has received mixed support and may not be equally viable across species and experimental conditions. Here we propose a novel intermediate strategy for approximating the quality of a detector and subject our own algorithm for spindle detection in dogs to the procedure. Automatic detections across the full range of our data (excluding only data sets with high subject overlap) were analyzed for their compliance with three face validity criteria, originally discovered by expert scorers in human EEG. These criteria are the shortness of the events (mostly below 2 s, maximum 6), the distribution of faster (> 13 Hz) spindles along the midline (higher in central and posterior derivations) and an association between amplitude and negative chirp (positive for sleep spindles and mostly unrelated for other oscillations). All three criteria were observed in the majority of data sets and sampling events (subsets of data defined by condition and/or recording channel). Replication was more consistent for data sets than for sampling events, but combined probabilities (Fisher’s method) always favoured face validity compliance. Our results indicate that automatic detections in the dog, which were shown in the past to display similar to humans associations with age and cognition, also share face validity criteria with human spindles. This strengthens the dog as a model in sleep spindle research and offers additional arguments for the quality of our detection method.
Implanted Brain-Computer Interfaces (BCIs) hold significant promise as a replacement for conventional assistive technologies for people with extensive motor impairments. In recent years, significant progress has been made in enhancing BCI performance, bringing clinically viable systems within reach. Some of these developments rely on increasingly large numbers of high spatial-density electrocorticography (ECoG) electrodes, which enable the extraction of spatially detailed information from extensive brain areas, but may also elevate surgical burden, posing a challenge for the clinical adoption of implanted BCIs. To mitigate this risk, we conducted an exhaustive investigation of hand movement classification performance involving 4, 5, and 8 classes in nine individuals with epilepsy, exploring all possible rectangular ECoG subgrids within the 32-, 64-, and 128-channel grids that were implanted in these individuals. Our findings reveal that the surface area of ECoG grids can be substantially reduced by 75–94
Graph-based white matter tractometry represents bundles as networks, preserving spatial topology that traditional along-tract profiles collapse. This enables the detection of distributed pathology patterns organized across connected regions. The field is at a critical juncture: methods are proliferating rapidly, but validation infrastructure lags. We systematically assess maturity across the analytical pipeline, from diffusion measurements through graph construction, detection models, clinical applications, and validation resources. Diffusion metrics measure water behavior rather than directly measuring tissue microstructure. Their biological interpretation relies on model assumptions that have been validated only in restricted contexts. Tractography provides spatial scaffolding with known limitations, and graph-construction choices encode implicit hypotheses about the organization of pathology but are often underspecified. Clinical studies demonstrate consistent group differences across disorders. Where controlled comparisons exist, graph methods show modest improvements over traditional approaches. However, comprehensive benchmarking against TBSS and AFQ is absent. Dedicated validation platforms for tractometry detection models do not yet exist. We document concrete barriers to systematic validation and assess emerging infrastructure addressing these gaps. Graph-based tractometry shows promise as a research tool. Realizing its clinical potential requires validation matching the maturity of traditional approaches.
Neuronal morphogenesis arises through coordinated neurite dynamics that generate cell-type specific dendritic branching during development. Recent advances in high-throughput time-lapse imaging techniques have transformed our ability to track such growth dynamics, yielding comprehensive anatomical datasets of neuronal morphologies. However, quantifying these structural trajectories during neuronal development remains a major challenge. We introduce the Temporal Topological Morphology Descriptor (TTMD), a framework that combines persistent topology with time-resolved neuronal imaging to quantify developmental trajectories of neuronal architecture. Applying TTMD to datasets of Drosophila sensory neurons, we show that developmental stages, branching processes, and mutant-specific growth dynamics can be decoded directly from topology without manual feature selection. Topological Morphology Descriptor (TMD) accurately resolves major developmental transitions in Class I Drosophila neurons, whereas temporal topology (TTMD) is required to uncover subtle alterations in branch dynamics caused by mutations in actin regulatory pathways in Class III Drosophila neurons. TTMD further enables automated topological tracking of branch emergence and retraction across development, overcoming the limitations of labor-intensive manual annotation. These results demonstrate that temporal topology provides a unified and interpretable language for neuronal morphogenesis, enabling the automated analysis of developmental neuroanatomy, and identifying pathological alterations in neuronal structure.
Computational neuroscience projects often combine simulation code, configuration files, analysis scripts, plotting utilities, and provenance records through loosely coupled and difficult-to-replay workflows. SACS is presented as a configuration-driven research-software framework for reproducible schematic multi-region circuit simulation with integrated analytics, validation checks, deterministic execution, deterministic replay, artifact replay, and graphical inspection. Implemented as the Python package brain_sim, the framework converts declarative model and scenario specifications into standardized run directories containing numerical outputs, machine-readable summaries, figure-generation recipes, validation reports, and provenance metadata. The contribution is methodological and neuroinformatics-oriented. SACS is not presented as a biologically validated model of anxiety, brain function, or treatment response, and its built-in circuit materials are used as configurable demonstration components rather than as evidence of clinical or biological validity. Instead, the framework is evaluated as software for reproducible computational experimentation: hypotheses are encoded as explicit configurations, executed under recorded seeds, inspected through analytics and validation layers, and then used to inform subsequent configurations in an iterative workflow. The manuscript describes the software architecture, configuration and execution model, artifact structure, replay mechanisms, and graphical/programmatic interfaces that support this workflow. The central claim is that SACS improves transparency, inspectability, and reuse for schematic circuit experiments by binding configuration, execution, analytics, provenance, and replay within a single software environment.
Brain-Machine Interfaces (BMIs) hold significant promise for the neurorehabilitation of patients with lower-limb impairments. However, their widespread clinical adoption is hindered by high costs and system complexity. This study presents an open-loop, low-cost EEG-based BMI designed to assess the cognitive implication of users during assisted cycling therapy. The system computes two cognitive indices: a low-frequency index, associated with motor-related engagement, and a high-frequency index, related to attention during motor tasks. These indices are obtained using Filter Bank Common Spatial Patterns (FBCSP) and Linear Discriminant Analysis (LDA), enabling continuous monitoring of cognitive involvement. Leave-One-Out Cross-Validation (LOOCV) results showed a clear increase in both indices during motor engagement compared to relaxed states in healthy subjects (low-frequency: 33.4 ± 7.1% to 65.1 ± 10.6% ; high-frequency: 28.1 ± 9.7% to 61.0 ± 11.2% ), whereas patients exhibited smaller increments (low-frequency: 39.9 ± 4.5% to 53.8 ± 10.3% ; high-frequency: 34.0 ± 4.2% to 48.0 ± 15.6% ). During validation trials, index differences between relax periods and motor engagement periods reveal a mean separation of 21.0 ± 16.3% (low-frequency) and 15.4 ± 7.9% (high-frequency) in control subjects, compared to 2.8 ± 7.6% and 2.4 ± 2.5% in patients. Results suggest that the metrics obtained by the present BMI can be a useful asset to evaluate patients’ performance, engagement and fatigue during neurorehabilitation.
Recent advancements in neural decoding have shown promising results in reconstructing visual experiences from brain activity. However, existing approaches focus primarily on decoding within a single dataset or subject, which limits generalization across various sources of neuroimaging. In this work, we propose a novel framework for the decoding of visual stimuli between subjects and between data sets, integrating neural recordings from multiple publicly available fMRI datasets. To address inherent intersubject and interdataset variability, we introduce a contrastive learning-based alignment strategy using image embeddings from a pre-trained IP-Adapter model. Our approach learns a shared latent space by aligning subject-specific neural representations with image features, enabling generalized decoding across both subjects and datasets. In addition, we propose a simple yet effective data augmentation method using ridge regression. This method synthesizes realistic fMRI-like signals from novel images by predicting voxel activity and injecting learned noise distributions, thus enhancing training diversity and model robustness. To the best of our knowledge, while several recent studies have explored cross-subject decoding, we extend recent cross-subject decoding efforts by training a single unified framework jointly across multiple public fMRI datasets and subjects, enabling cross-dataset transfer in addition to cross-subject generalization. We distinguish this multi-dataset unified training setting, where each dataset contributes training data, from a stricter leave-one-dataset-out transfer setting in which the target dataset is excluded from source pretraining and used only for lightweight alignment-layer adaptation. Empirically, our unified model achieves strong semantic reconstruction across datasets (e.g., up to 94.8
Quantitative analysis of neuronal morphology in dense cultures remains technically challenging, as extensive overlap between neuritic arbors and background signal limits the applicability of traditional tracing-based approaches. Here, we applied fractal dimension (FD) and lacunarity (LAC) analysis as scale-invariant, global measures of neurite complexity and systematically compared them with classical morphometric approaches in dense primary hippocampal neuronal cultures. Neurons were treated with the Abl tyrosine kinase inhibitor imatinib to induce cytoskeletal remodeling. Fractal analysis was performed on skeletonized representations derived from standardized binarized confocal images, enabling robust quantification of neurite space-filling properties across spatial scales. Imatinib treatment induced a pronounced reduction in FD (DMSO: 1.56 ± 0.04 vs. IMA: 1.36 ± 0.05, p < 0.001) together with increased lacunarity, indicating reduced multiscale complexity and increased spatial heterogeneity. In contrast, classical morphometric analysis revealed a heterogeneous phenotype, with elongation of primary neurites accompanied by reduced branching. These opposing effects partially compensated for one another, obscuring overall complexity changes when assessed using individual parameters. To integrate multivariate features, we constructed a composite Neurite Complexity Index (NCI) incorporating branching, neurite length, and distribution-sensitive descriptors (R² ≈ 0.98; F(4,106) = 89.6, p < 1 × 10⁻184). FD showed strong convergence with the NCI across bootstrap analyses, and clustering and discriminant analyses confirmed its high discriminatory power. Together, these findings suggest that fractal analysis may provide a sensitive and compact framework within the present experimental context for detecting treatment-associated structural changes in dense neuronal cultures.
Positron Emission Tomography (PET) diagnostic precision is often compromised by low spatial resolution. Deep learning restoration models tend to sacrifice quantitative accuracy for visual sharpness, and most are trained on a single fixed degradation profile, limiting generalization across scanners. This paper presents a metabolically faithful 3D restoration framework pairing a volumetric extension of SwinFIR with two innovations: (1) a composite metabolic-aware loss enforcing structural, distributional, and frequency-domain agreement with the ground truth, and (2) a stochastic degradation augmentation strategy that randomizes point spread function parameters, voxel sampling, and counting noise during training, exposing the model to a distribution of simulated scanner-like degradations rather than a single fixed simulation. Evaluated on NeuroEXPLORER data, the proposed method outperforms baselines with a Structural Similarity Index Measure (SSIM) of 0.843, Peak Signal to Noise Ratio (PSNR) of 27.08 dB, and Normalized Root Mean Squared Error (NRMSE) of 0.117, while maintaining metabolic fidelity (Concordance Correlation Coefficient (CCC) 0.948, Wasserstein distance 0.018). Ablation experiments confirm that stochastic degradation augmentation improves robustness over fixed-profile training. The framework recovers anatomical detail with only small but measurable regional SUVR biases ( ≤ 3.3
Functional connectivity (FC) is a widely used metric in functional magnetic resonance imaging (fMRI) research. However, its reliability has long been a notable concern, especially in studies with small sample sizes. Previous research has demonstrated that FC derived from longer fMRI scans exhibits higher reliability, making the prediction of long-scan FC from existing short-scan FC data a feasible and promising approach. First, we constructed three general linear models (GLMs) to predict long-scan FC from short-scan FC data using the Human Connectome Project (HCP) dataset. Next, we interpreted the models by visualizing their weight distributions. Subsequently, we validated our findings across multiple independent datasets and with different machine learning models. Finally, we applied the models to enhance both the test-retest reliability of FC and the performance of connectome-based predictive modeling (CPM). Our results showed that GLMs based on individual short-scan FC successfully predicted individual long-scan FC values. Moreover, the differences between the three GLMs could be explained by the distinct distribution characteristics of the FC matrices they predicted. Our findings were validated using data from the Consortium for Reliability and Reproducibility (CoRR) project and an in-house local dataset. Additionally, our models outperformed conventional machine learning approaches. Critically, these models effectively improved both the test-retest reliability of FC and the predictive performance of CPM. In conclusion, GLMs built on individual short-scan FC can robustly predict individual long-scan FC values. These models show strong generalizability across different datasets, and can be widely applied to improve the test-retest reliability of FC and the performance of CPM in neuroimaging studies.
Computational morphometry has transformed the quantitative analysis of peripheral nerve structure, enabling large-scale, computational, and longitudinal studies that were previously impractical using manual methods. However, this review argues that the reliability and interpretability of morphometric outputs are fundamentally pipeline-conditional, shaped by assumptions introduced across sample acquisition, preparation, imaging, annotation, segmentation, and metric extraction rather than by segmentation accuracy alone. By examining the full morphometry pipeline, we show how protocol variability, limited model generalization, and ambiguity in expert-defined ground truth propagate downstream and constrain reproducibility, particularly in pathological tissue. Using peripheral nerve morphometry as a tractable model system, we highlight issues that are representative of broader challenges in medical image analysis and quantitative neuroanatomy. We conclude that progress in computational morphometry will depend less on incremental algorithmic improvements and more on shared datasets, uncertainty-aware validation, and closer alignment between structural metrics and functional relevance in both experimental and clinical contexts.
Dynamic functional connectivity (dFC) analysis in functional magnetic resonance imaging (fMRI) faces a fundamental challenge: conventional sliding-window methods must trade temporal resolution against statistical reliability, while rare transient neural events risk becoming undetectable when included in training data. We introduce HESREN (Hermite-Enhanced Software Reservoir Network), a novel framework integrating echo state networks with derivative-informed Hermite-type neural operators to enable windowless dFC estimation and leakage-free transient detection. HESREN employs a leaky-integrator reservoir that projects multivariate fMRI time series into high-dimensional state spaces, augmented with Gaussian-smoothed temporal derivatives to form enhanced feature vectors encoding value, velocity, and acceleration. Strict temporal partitioning trains all components exclusively on baseline segments while evaluating on complete time series, preserving transient events as out-of-distribution signals. Teacher-student distillation transfers the temporal precision of micro-window connectivity estimates into stable windowless operators via ridge-regularised linear readout; all hyperparameters and initialisation procedures are fully specified to ensure reproducibility. Validation on the NEBULA101 resting-state fMRI dataset across N=50 participants demonstrates consistent and substantial improvements over conventional methods. Transient event detection achieves AUC =0.881± 0.025 and average precision AP =0.938± 0.018 , compared to AUC =0.677± 0.074 for raw-derivative baselines (Wilcoxon W=1275 , p<0.0001 , Cohen’s d=2.84 ), with phase-randomised surrogate testing confirming statistical robustness in all participants ( p=0.005 , n=200 surrogates). Comparison against mainstream dFC alternatives shows that HESREN statistically significant performance gains Gaussian Hidden Markov Models (AUC =0.805± 0.088 ), temporal convolutional networks (AUC =0.815± 0.062 ), LSTM autoregressive predictors (AUC =0.580± 0.080 ), and conventional sliding-window correlation (AUC =0.690± 0.061 ), with all advantages statistically significant ( p≤ 0.028 ). Windowless dFC trajectories attain lag-corrected correlation r_lag=0.256± 0.019 with micro-window teachers while providing 3– 5× finer temporal resolution than 25-TR sliding windows. Network-level analysis reveals that HESREN detects transient events an average of 4.5 TR (9 s) earlier than sliding-window methods, selectively amplifies within-language-network coupling by 97% and default-mode-network coupling by 119% during detected events, and is the only evaluated method to yield a positive network segregation index ( SI=+0.038 ), consistent with the known modular organisation of resting-state brain networks. HESREN overcomes fundamental limitations of sliding-window dFC through derivative-aware reservoir dynamics, offering a computationally efficient, mathematically principled framework for capturing transient neural reconfigurations with temporal precision previously improved in fMRI connectivity analysis. The modular architecture facilitates adaptation to diverse neuroimaging applications, from basic neuroscience to real-time clinical monitoring systems.
This study develops a fractional-order model of Alzheimer's disease using a [Formula: see text]-generalized Atangana-Baleanu-Caputo (ABC) operator to capture the spatiotemporal dynamics of amyloid-beta and tau protein spread, coupled with a neuron regeneration mechanism. The fractional parameters α, [Formula: see text], and τ control memory depth, deformation of the kernel, and temporal scaling, respectively. Numerical simulations demonstrate that: (i) intermediate fractional orders [Formula: see text] produce biologically realistic propagation delays, (ii) lower [Formula: see text] values enhance nonlocal interactions and accelerate tau diffusion across the connectome, and (iii) increasing the scaling parameter τ slows accumulation, mimicking effective clearance or treatment response. Incorporating a treatment term with drug diffusion and decay reveals that sustained low decay rates ([Formula: see text]) markedly reduce tau concentrations and protect neuron populations. These findings show that the [Formula: see text]-ABC framework not only captures the hereditary and memory effects of Alzheimer's progression but also provides a flexible platform for simulating therapeutic interventions and predicting disease trajectories using real brain connectome data.
This study develops a fractional-order model of Alzheimer’s disease using a (q,τ ) -generalized Atangana-Baleanu-Caputo (ABC) operator to capture the spatiotemporal dynamics of amyloid-beta and tau protein spread, coupled with a neuron regeneration mechanism. The fractional parameters α , q , and τ control memory depth, deformation of the kernel, and temporal scaling, respectively. Numerical simulations demonstrate that: (i) intermediate fractional orders 0.6 ≤α≤ 0.9 produce biologically realistic propagation delays, (ii) lower q values enhance nonlocal interactions and accelerate tau diffusion across the connectome, and (iii) increasing the scaling parameter τ slows accumulation, mimicking effective clearance or treatment response. Incorporating a treatment term with drug diffusion and decay reveals that sustained low decay rates ( μ < 0.3 ) markedly reduce tau concentrations and protect neuron populations. These findings show that the (q,τ ) -ABC framework not only captures the hereditary and memory effects of Alzheimer’s progression but also provides a flexible platform for simulating therapeutic interventions and predicting disease trajectories using real brain connectome data.
Many functional properties vary dramatically across neurons in cerebral cortex. Two fundamental goals of systems neuroscience are to determine which neurons execute which functions and how the different functional properties of a neuron are related. Here we focus on functional segregation - when two separate subpopulation of neurons encodes two distinct functions. Often, it is assumed that if two functional properties are uncorrelated across the population, then there is no functional segregation. Here we show that this assumption can lead to wrong conclusions; functional segregation can emerge, by chance, due to random variation when that variation is distributed according to skewed, heavy-tailed distributions. We reexamine the results we previously reported (Nature communications, 10(1), 1575-1575 2019), which showed that neurons in primary motor cortex tend to be functionally segregated, with neurons that are either strongly coupled to body movements or to ongoing cortical population activity. Here we show that this is a prime example of functional segregation due to random variation.
Cognition under uncertainty can be formalized through Bayesian inference, but biologically plausible neural implementations remain a challenge. This study develops a Bayesian neural model for lifespan prediction that integrates fractional-order dynamics into classical Leaky Integrate-and-Fire and Izhikevich neuron models. The inclusion of fractional derivative introduces long-term memory, and thus enhancing both biological plausibility and representational capacity of the Bayesian neural model. Experimental results demonstrate that fractional-order neuron models consistently provide closer alignment with both human predictions and optimal Bayesian predictions. The large-scale fractional-order Izhikevich model shows the most robust convergence and cortical plausibility. These findings highlight the role of fractal neural dynamics in probabilistic cognition and bridging theoretical Bayesian models with realistic spiking behavior. The study demonstrates how biologically inspired spiking neuron models can approximate Bayesian inference, suggesting pathways for computational neuroscience to design models that learn, predict, and adapt with the efficiency of cortical computations. Further, in this study, neural populations represent priors from demographic lifetables. However, a uniform likelihood and posterior that yield median lifespan predictions as probability distributions within the Neural Engineering Framework have been retained from the previous study. To investigate the influence of neural population size on biological plausibility and Bayesian optimality, experimental conditions systematically increase the neural population size and compare the predictive outcomes.
One of the main objectives of cognitive neuroscience is to investigate brain processes that underlie narrative comprehension. Furthermore, earlier studies that used naturalistic functional magnetic resonance imaging (fMRI) datasets, like Narratives, has advanced our knowledge of large-scale language and narrative networks, most studies have relied on correlation-based analyses or single-region importance measures, overlooking the dynamic and structural properties of brain networks. In this work, we present a new graph-based framework to identify important regions in narrative comprehension by combining a composite node importance scoring method with multiple node embedding algorithms. We first used controlled simulations with stochastic block models (SBM) with different hub nodes and community strengths to validate the framework. This made it possible to systematically assess seven embedding algorithms for node influence attribution, link prediction, and community detection. Applying the same framework to fMRI data, we analyzed two parcellation schemes, the Harvard-Oxford and Schaefer (100-parcel) atlases, to identify influential cortical regions. Our findings reveal consistent engagement of the default mode, salience, and limbic networks across stories and atlases, emphasizing their central role in narrative processing. Overall, this work offers a reliable, comprehensible method for identifying key brain regions, bridging the gap between graph representation learning and cognitive neuroscience. The framework provides a scalable basis for further research that connects naturalistic cognition, dynamic brain connectivity, and linguistic features.
