Objective.Local field potential (LFP) decoding is critical for the clinical translation of intracortical brain-machine interfaces, yet existing decoding methods are limited by three key bottlenecks: insufficient single-scale feature utilization, inefficient multi-scale feature fusion, and poor robustness across task paradigms and chronic recording conditions.Approach.To address these challenges, we propose Dual-VCT, a novel dual-branch variational mode decomposition-convolutional neural network-Transformer (VMD-CNN-Transformer) model for end-to-end LFP decoding. The core innovation of Dual-VCT is its symmetric time-frequency parallel architecture with independent VMD modules embedded in both branches: a temporal branch decomposes local motor potential (LMP) signals via VMD to capture motion-related instantaneous neural activity, while a frequency-domain branch leverages VMD to isolate task-relevant spectral power components, with a hierarchical fusion pipeline enabling robust cross-scale feature integration.Main results.Validated in non-human primate experiments, Dual-VCT achieved a classification accuracy of 0.930 ± 0.023 in the 3-class spatial grasping task, and a Pearson correlation coefficient (CC) of 0.910 ± 0.023 in the finger point-to-point tracking task. It significantly outperformed all comparative dual-branch methods under identical experimental conditions (p< 0.05), delivered a 4% performance gain over single-feature decoding, and exhibited strong cross-task robustness and cross-day stability. Ablation experiments confirmed the core contribution of the dual-branch VMD design.Significance.This work provides a high-performance structured paradigm for LFP decoding, with a clinically oriented design that supports the long-term stability of chronic iBMI systems.

Although the broad learning system (BLS) and its existing robust variants have been widely applied in various fields due to their excellent performance, they still cannot effectively handle noise present in the input data of training samples, which may lead to a decline in algorithm performance. To address this issue, this paper proposes a maximum total correntropy-based BLS (MTC-BLS), which trains the model using the maximum total correntropy (MTC) criterion. This endows the model with the advantages of both the total least squares (TLS) method and the MCC criterion, enabling it to effectively handle both input and output noise. Furthermore, to mitigate the impact of kernel width selection deviations in the entropy-based criterion, two new robust methods are further proposed, namely MMTC-BLSa and MMTC-BLSb. The former directly incorporates the M-estimator into the algorithm’s objective function, using the dual constraints of the M-estimator and MTC to balance model predictive performance while reducing the algorithm’s dependence on the kernel width. The latter uses the M-estimator as a weighting factor in the MTC criterion for model training, employing the M-estimator to compensate for model errors caused by kernel width deviations, thereby effectively mitigating the negative effects of such deviations. Moreover, to enable efficient training, fixed-point iteration methods are introduced to provide iterative solution strategies for MTC-BLS, MMTC-BLSa, and MMTC-BLSb, respectively. The effectiveness and superiority of the proposed methods are validated through multiple comparative experiments on time series datasets, regression datasets and image datasets.

Interactions between non-coding RNAs and proteins (NPIs) are pivotal in diverse biological processes and disease mechanisms. While computational methods utilizing molecular structure and graph topology are widely used for NPI prediction, they often struggle to achieve higher accuracy due to their dependence on prior knowledge and their inability to capture correlation information between NPI sequences. In this paper, we introduce a sequence-based NPI prediction model that integrates Longformer with contrastive learning. Our model utilizes Longformer to extract features from long biological sequences and employs the representation-based method for feature interaction. Furthermore, the InfoNCE loss is applied during training to enhance the quality of feature embeddings. Our model achieved accuracies of 88.7%, 98.0%, and 95.1% on the RPI488, RPI1807, and NPInter2.0 datasets, respectively. Experiments demonstrate that our model is comparable to or better than several state-of-the-art NPI prediction methods on both large-scale and small-scale datasets. This superior performance validates the effectiveness of our proposed model.

Motor imagery (MI) is a classical paradigm in brain-computer interfaces (BCIs) that relies on electroencephalogram (EEG) signals. While deep learning has significantly improved MI EEG decoding performance, most existing models are purely data-driven and overlook valuable prior knowledge. This paper proposes a knowledge-data fusion network (KDFNet), which integrates MI-specific neurophysiological priors with a data-driven convolutional neural network. Specifically, temporal and spatial layers are respectively initialized using band-pass and common spatial filters to integrate sensory-motor rhythms and event-related desynchronization/synchronization patterns. A log-variance activation further embeds power spectrum information, while the network’s output layer is initialized using parameters from a traditional classifier to align its decision boundary with classical decoding strategies. Then, the model is further optimized via gradient descent to reduce the classification loss, enabling it to adaptively refine representations based on task-specific data. Experiments on four MI datasets demonstrated that KDFNet outperforms competitive baselines, achieving an average improvement of 3.07 % in within-subject classification and 2.93 % in cross-subject classification. This work demonstrates the benefits and necessity of integrating prior knowledge with data-driven models in EEG-based BCIs.

Alzheimer's disease (AD) is a prototypical neurodegenerative disorder characterized by dynamic structural changes in the brain as the disease progresses. Longitudinal magnetic resonance imaging (MRI) can capture structural alterations at different time points, but data gaps during certain intervals of clinical follow-up often limit continuous analysis of disease progression. To address this challenge, this study proposes an MRI temporal interpolation method based on bidirectional deformation modeling. This approach models the generation of MRI images at intermediate time points as an image interpolation task while leveraging longitudinal temporal information. Specifically, it employs a multiscale optical flow estimation network to predict bidirectional deformations between consecutive MRI scans. It combines adaptive masks with a CBAM attention mechanism to enhance modeling of structural changes in critical brain regions. Furthermore, a shared feature encoder extracts multi-level contextual information, while teacher-student knowledge distillation further improves generation quality. Experiments validated the model's performance on gray matter, and white matter images through midpoint interpolation and continuous temporal interpolation tests. The model was pretrained on the Vimeo90K video dataset and evaluated on the ADNI1/ADNI2 longitudinal MRI datasets. Results demonstrate that the proposed method generates structurally coherent and temporally continuous intermediate MRI images, providing an effective imaging tool for the early diagnosis and ongoing assessment of Alzheimer's disease (AD) and mild cognitive impairment (MCI).

Decoding brain signals accurately and efficiently is crucial for intra-cortical brain-computer interfaces. Traditional decoding approaches based on neural activity vector features suffer from low accuracy, whereas deep learning-based approaches have high computational cost. To improve both the decoding accuracy and efficiency, this paper proposes a multi-neuron spiking neural network (MnSNN) for effective and energy-efficient intra-cortical brain signal decoding. We also propose a feature fusion approach, which integrates the manually extracted neural activity vector features with those extracted by a deep neural network, to further improve the decoding accuracy. Experiments in decoding motor-related intra-cortical brain signals of two rhesus macaques demonstrated that our MnSNN achieved higher accuracy than traditional artificial neural networks; more importantly, it was tens or hundreds of times more efficient than traditional artificial neural networks. The MnSNN model is very suitable for high precision and low power applications like intra-cortical brain-computer interfaces.

Broad learning systems (BLSs) have been widely used in regression and classification because of their simple network structure and high computational efficiency. However, for time-series forecasting tasks, existing BLS models and their variants still have difficulty fully exploiting temporal information, especially in three-dimensional (3D) and four-dimensional (4D) time-series forecasting. To overcome this limitation, this paper proposes a quaternion broad learning system (NQBLS). The proposed model constructs quaternion-valued inputs and uses split activation functions to generate network neurons. In this way, NQBLS can capture latent relationships among different output variables in time-series data and encode temporal information for complex time-series patterns. Comprehensive comparative experiments are conducted to verify the effectiveness of NQBLS on 3D/4D time-series forecasting tasks.

Objective: A non-invasive brain-computer interface (BCI) enables direct interaction between the user and external devices, typically via electroencephalogram (EEG) signals. This paper tackles the problem of decoding EEG signals across different headsets, which is challenging due to differences in the number and locations of the electrodes. Methods: We propose a spatial distillation based distribution alignment (SDDA) approach for heterogeneous cross-headset transfer in non-invasive BCIs. SDDA uses first spatial distillation to make use of the full set of electrodes, and then input/feature/output space distribution alignments to cope with the significant differences between the source and target domains. Results: Extensive experiments on six EEG datasets from two BCI paradigms demonstrated that SDDA achieved superior performance in both offline unsupervised domain adaptation and online supervised domain adaptation scenarios, consistently outperforming 10 classical and state-of-the-art transfer learning algorithms. Significance: Our approach enables effective transfer between heterogenous EEG headsets, improving and expediting BCI calibration.

For intracortical brain-machine interfaces (iBMIs), a critical bottleneck in clinical translation is the insufficient decoding performance and poor generalization of spike-local field potential (LFP) fusion methods, which stems from the lack of artificial intelligence (AI) designs tailored to neural signal heterogeneity. This study proposes a lightweight AI-driven multi-scale fusion neural decoding (MFND) framework, integrating deep learning and self-attention mechanisms, for high-precision motor intention decoding in iBMIs. Distinct from state-of-the-art (SOTA) single-stage model-driven multimodal fusion methods, MFND adopts a three-branch hierarchical fusion architecture for AI-based heterogeneous neural signal processing. Validated on two non-human primate iBMI datasets, MFND achieves a classification accuracy of 0.93 ± 0.025 (2% absolute improvement over SOTA) in three-class motor tasks, and a Pearson's correlation coefficient (CC) of 0.91 ± 0.017 (2% absolute improvement over SOTA) in movement trajectory regression. Cross-day validation yields 2% higher accuracy and 4% higher CC than SOTA, with statistically significant differences (paired Wilcoxon signed-rank test, p < 0.05). With open-source code ensuring reproducibility, this compact framework provides a feasible solution for high-precision, long-term stable iBMI decoding, and facilitates AI-driven clinical translation in neuroengineering.

Background Intracortical brain-machine interfaces (iBMI) hold promise for restoring lower limb mobility in patients with spinal cord injury (SCI). Although spike-based signals have been widely used for locomotion decoding, local field potentials (LFPs) offer a complementary signal source for investigating and decoding SCI-related hindlimb movement. Methods Using a rodent spinal cord injury model (n=3) with motor cortex implants of 16-channel microwire electrode, we integrated bipedal treadmill training and T10 hemisection to compare pre-/post-spinal-cord-injury neural dynamics. Neural signals and hindlimb kinematic data were simultaneously recorded. Linear Discriminant Analysis (LDA)、Support Vector Machine (SVM) and Fully Connected Neural Network (FCNN) based on the low-frequency bands of the LFP signals were used to decode the foot-off events during bipedally walking tasks both before and after spinal cord injury. Moreover, the characteristics of the LFP signals before and after spinal cord injury were explored and compared. Results Comparing the decoded performance on the pre-/post-spinal-cord-injury, all the models showed the enhanced F1-scores post-spinal cord injury: LDA (+37.1% to 0.617), SVM (+1.8% to 0.629) and FCNN (+38.1% to 0.649). Moreover, low-frequency LFP power showed apparent phase-related concentration across recovery stages, whereas higher-frequency LFP power showed weaker clustering, with a transient tendency toward concentration during the short-term recovery phase. Conclusion This demonstrates compensatory cortical control and establishes local-field-potential-based decoding as a viable solution when spike signals degrade. The findings advance stable intracortical brain-machine interface development for spinal cord injury rehabilitation. Funding This work was supported in part by the National Natural Science Foundation of China (82572358), Key Research and Development Program of Wuhan (2025061202030424), Nanjing Major Science and Technology Special Project (202512138), and Interdisciplinary research program of HUST 2025JCY064.

Monetary, fiscal, and credit interventions are macro-financial regulation policies and are critical to ensuring financial stability. The nonlinear, high-dimensional, and uncertain macroeconomic and economic systems make it difficult to measure their effectiveness. The research builds a new Bi-directional Long Short-Term Memory with Multi-Head Attention Transformer tuned Takagi–Sugeno Fuzzy Network (Bi-LSTM-MHAT-TSFN) to deliver accurate, robust, and interpretable macro-financial policies evaluation. Pre-processed through normalization, missing-value handling, outlier filtering, and autoencoder-based dimensionality reduction, the input comprised a hybrid macro-financial dataset combining publicly available indicators with synthetically generated samples, and then dimensionality reduced in 2019–2025 Gross Domestic Product (GDP) growth, inflation, interest rates, credit, leverage, asset prices, and policy interventions. A Bi-LSTM network that could find patterns, lags, and structural shocks in macro-financial sequences was used to capture temporal dependencies. The MHAT module improved learning across variables by making crucial connections more obvious. The TSFN layer then turned these improved features into clear, understandable fuzzy rules that made it easier to evaluate policies. All implementation procedures were carried out using Python. The Bi-LSTM-MHAT-TSFN model showed an accuracy of 98.8

Accurate remaining useful life (RUL) prediction is critical for aeroengine prognostics and health management. However, standard Transformer-based models struggle to capture multiscale temporal dynamics and cross-sensor dependencies in aeroengine sensors. Moreover, time-domain scaled dot-product attention does not exploit the degradation characteristics of onboard sensor series or the low-rank structure evident in the frequency domain. We propose Multi-sensor Spatial and Multiscale Temporal Transformer guided by Frequency-Domain Attention (MSMT-FDA), which couples a dual-branch encoder-encompassing multiscale temporal cues and multi-sensor spatial correlations-with frequency-domain low-rank distance attention, leveraging spectral low-rank priors of engine degradation signals. On C-MAPSS (FD001-FD004) and DS02 of N-CMAPSS, our method attains state-of-the-art performance and remains robust across flight regimes, enabling predictive maintenance.

As a generalization of covering, fuzzy /3-covering provides a more accurate and practical representation for incomplete information. This paper primarily proposes several fuzzy neighborhood operators based on diverse aggregation functions in an fuzzy /3-covering approximation space (F/3CAS) and develops a novel TOPSIS method to address the decision-making problem related to user preference factors. First, two classes of fuzzy neighborhood operators are introduced, derived from t-norms, overlap functions and their residual implications in an F/3CAS, with their properties thoroughly analyzed. In addition, multiple fuzzy /3-coverings are generated from the original fuzzy /3-covering, and the classifications of fuzzy neighborhood operators, along with their partial order relationships, are examined. Based on these operators, two kinds of fuzzy /3-covering-based rough sets (F/3CRS) are established. Finally, an F/3CRS-based fuzzy TOPSIS method is developed to evaluate user preference factors for fresh fruit, thereby demonstrating the rationality and feasibility of the proposed approach.

Intracortical brain-machine interfaces (iBMIs) aim to establish a communication path between the brain and external devices. However, in the daily use of iBMIs, the non-stationarity of recorded neural signals necessitates frequent recalibration of the iBMI decoder to maintain decoding performance, which requires collecting and labeling a large amount of new data. To address this challenge and minimize the time needed for decoder recalibration, we proposed an active learning domain adversarial neural network (AL-DANN). This model leveraged a substantial volume of historical data alongside a small amount of current data (four samples per category) to calibrate the decoder. By incorporating domain adversarial and active learning strategies, the model effectively transferred knowledge from historical data to new data, reducing the demand for new samples. We validated the proposed method using neural signals recorded from three monkeys performing different movements in a classification task or a regression task. The results showed that the AL-DANN outperformed existing state-of-the-art methods. Impressively, it required only four new samples per category for decoder recalibration, leading to a recalibration time reduction of over 80 %. To our knowledge, this is the first study to incorporate deep transfer learning into iBMI decoder calibration, highlighting the significant potential of applying deep learning technologies in iBMIs.

As an extension of partition, fuzzy β-covering can provide a more realistic and accurate description of incomplete information. In this paper, we mainly integrate the idea of fuzzy β-covering with dual hesitant fuzzy (DHF) information and construct some novel three-way decision (3WD) models with DHF covering-based rough set. Firstly, we propose the notions of DHF β-covering, DHF β-minimal description and DHF β-maximal description, and then construct three types of DHF neighborhood operators. Meanwhile, we introduce DHF conditional probability by using DHF neighborhood operator. Secondly, in terms of DHF conditional probability, DHF covering-based probabilistic rough set, DHF covering-based decision-theoretic rough set and their 3WD models are established. In light of DHF neighborhood operator, we propose two pairs of optimistic and pessimistic DHF decision evaluation functions and build two 3WD models through them. Lastly, a numerical example is employed to elaborate the application of the above models, which is effective and credible to medicine screening for Alzheimer’s disease.

The brain, recognized as one of the most intricate systems globally, has been a focal point for scientific exploration. Researchers have made efforts to construct models of the brain based on neural dynamics and complex networks to gain insights into its workings. It is crucial to investigate the brain's working principles from various perspectives. This study presents a novel thermophysical model of the motor cortex and examines its potential thermodynamic properties. Utilizing canonical ensemble theory, we constructed the thermophysical model using spike and local field potential (LFP) signals obtained from intracortical brain-machine-interfaces (iBMIs) in two monkeys. The parameters derived from this model—namely internal energy, free energy, and entropy—were employed to assess the thermodynamic properties and observe alterations in these properties during reaching and grasping movements. Furthermore, this proposed model was applied to movement pattern decoding, highlighting its potential in neural decoding tasks. In both LFP- and spike-based thermodynamic models, there was an increase in internal energy and free energy, coupled with a decrease in entropy when the motor cortex was activated across various movement tasks. This suggests that the neural system adheres to the principles of a thermophysical system. Notably, the thermodynamic features demonstrated superior performance in decoding movement intentions compared to traditional LFP and spike features. This study represents the first construction of a comprehensive thermodynamic model of the motor cortex based on LFP and spike signals. The model exhibits remarkable stationarity and holds promise for long-term and stable evaluations of motor cortex functions.

ObjectiveThe present work concentrated on validating whether sinomenine alleviates bleomycin (BLM)-induced pulmonary fibrosis, inflammation, and oxidative stress.MethodsA rat model of pulmonary fibrosis was constructed through intratracheal injection with 5 mg/kg BLM, and the effects of 30 mg/kg sinomenine on pulmonary inflammation, fibrosis, apoptosis, and 4-hydroxynonenal density were evaluated by hematoxylin and eosin staining, Masson's trichrome staining, TUNEL staining, and immunohistochemistry. Hydroxyproline content and concentrations of inflammatory cytokines and oxidative stress markers were detected using corresponding kits. MRC-5 cells were treated with 10 ng/ml PDGF, and the effects of 1 mM sinomenine on cell proliferation were assessed by EdU assays. The mRNA expression of inflammatory cytokines and the protein levels of collagens, fibrosis markers, and key markers involved in the TLR4/NLRP3/TGF beta signaling were tested with RT-qPCR and immunoblotting analysis.ResultsSinomenine attenuated pulmonary fibrosis and inflammation while reducing hydroxyproline content and the protein expression of collagens and fibrosis markers in BLM-induced pulmonary fibrosis rats. Sinomenine reduced apoptosis in lung samples of BLM-challenged rats by increasing Bcl-2 and reducing Bax and cleaved caspase-3 protein expression. In addition, sinomenine alleviated inflammatory response and oxidative stress in rats with pulmonary fibrosis induced by BLM. Moreover, sinomenine inhibited the TLR4/NLRP3/TGF beta signaling pathway in lung tissues of BLM-stimulated rats. Furthermore, TLR4 inhibitor, TAK-242, attenuated PDGF-induced fibroblast proliferation and collagen synthesis in MRC-5 cells.ConclusionSinomenine attenuates BLM-caused pulmonary fibrosis, inflammation, and oxidative stress by inhibiting the TLR4/NLRP3/TGF beta signaling, indicating that sinomenine might become a therapeutic candidate to treat pulmonary fibrosis.

Fuzzy rough set (FRS) has a great effect on data mining processes and the fuzzy logical operators play a key role in the development of FRS theory. In order to further generalize the FRS theory to more complicated data environments, we firstly propose four types of fuzzy neighborhood operators based on fuzzy covering by overlap functions and their implicators in this paper. Meanwhile, the derived fuzzy coverings from an original fuzzy covering are defined and the equalities among overlap function-based fuzzy neighborhood operators based on a finite fuzzy covering are also investigated. Secondly, we prove that new operators can be divided into seventeen groups according to equivalence relations, and the partial order relations among these seventeen classes of operators are discussed, as well. Go further, the comparisons with $ t$-norm-based fuzzy neighborhood operators given by D'eer et al. are also made and two types of neighborhood-related fuzzy covering-based rough set models, which are defined via different fuzzy neighborhood operators that are on the basis of diverse kinds of fuzzy logical operators proposed. Furthermore, the groupings and partially order relations are also discussed. Finally, a novel fuzzy TOPSIS methodology is put forward to solve a biosynthetic nanomaterials select issue, and the rationality and enforceability of our new approach is verified by comparing its results with nine different methods.

Objective: Smoking has been suggested as a modifiable and cardiovascular risk factor for chronic kidney disease (CKD). Although long-term smoking has been associated with CKD, the potential relationship between its metabolite hydroxycotinine and CKD has not been clarified. Methods: A total of 8,544 participants aged 20 years and above from the National Health and Nutrition Examination Survey (NHANES) 2017 - March 2020 were enrolled in our study. CKD was defined by estimated glomerular filtration rate (eGFR) < 60 mL/(min*1.73 m(2)). Serum hydroxycotinine was measured by an isotope-dilution high-performance liquid chromatography/atmospheric pressure chemical ionization tandem mass spectrometric (ID HPLC-APCI MS/MS) method with a lower limit of detections (LLOD) at 0.015 ng/mL. The non-linear relationship was explored with restricted cubic splines (RCS). Pearson's correlation coefficient and a multivariate logistic regression model were used for correlation analysis. Results: Serum hydroxycotinine and eGFR were negatively correlated in both non-CKD group (r= -0.05, p < 0.001) and CKD group (r= -0.04, p < 0.001). After serum hydoxycotinine dichotominzed with LLOD, serum hydroxycotinine >= 0.015 ng/mL was negatively correlated with eGFR not only in non-CKD group (r = -0.05, p < 0.001) but also in CKD group (r = -0.09, p < 0.001). After adjusting for comprehensive confounders, results from the multivariate logistic regression analysis showed that participants with serum hydroxycotinine >= 0.015 ng/mL had increased odds of CKD (OR = 1.505, p < 0.001). Conclusions: Serum hydroxycotinine might be positively associated with CKD. Further study is warranted to find the right concentration of hydroxycotinine to measure the CKD.