Parkinson's disease (PD) diagnosis remains challenging because subtle neural alterations may be difficult to capture using conventional clinical assessment alone. This study proposes an attention-based deep learning framework for classifying PD from resting-state EEG with minimal preprocessing and leakage-safe evaluation. Raw EEG recordings were first partitioned at the subject level. Within each fold, the selected motor-related EEG channel was decomposed into canonical sub-bands using discrete wavelet transform, and the resulting sub-band signals were then segmented into overlapping temporal windows. Each sub-band window was transformed into a time-frequency spectrogram using the short-time Fourier transform and classified using a ResNet-101 backbone enhanced with dual channel-spatial attention. Hyperparameters were optimized using an Enhanced Adaptive Hybrid Covariance Matrix Adaptation Evolution Strategy (AH-CMA-ES), applied only within training/internal-validation subjects in each fold. Model performance was evaluated on two independent public EEG datasets, UC San Diego and University of Iowa, using subject-wise nested leave-one-subject-out cross-validation. In each fold, the held-out subject was excluded from training, augmentation, hyperparameter optimization, early stopping, and model selection. The proposed framework achieved 95.2% segment-level and 96.77% subject-level accuracy on UCSD, and 92.1% segment-level and 92.86% subject-level accuracy on Iowa, with subject-level decisions obtained by majority voting over non-augmented test segments. In addition to classification, sub-band topographical analysis provided exploratory neurophysiological interpretation across canonical EEG rhythms, revealing patterns consistent with reported PD-related oscillatory alterations. These findings suggest that resting-state EEG combined with attention-based deep learning can support robust, interpretable PD classification, while larger heterogeneous cohorts are needed to further validate clinical generalizability.

Accurate and efficient neuronal reconstruction is essential for large-scale neuronal projection analysis and neural circuit mapping. However, conventional reconstruction approaches are often constrained by the structural complexity of neurons, the diversity of imaging signals, and variations in annotator expertise, making it difficult to simultaneously achieve high reconstruction quality and efficiency. To address these challenges, this study proposes a quantitative and precision-oriented neuronal reconstruction framework that systematically integrates reconstruction efficiency and accuracy modeling, data-algorithm matching, and refined task allocation strategies. First, mathematical models were established to quantitatively characterize reconstruction efficiency and accuracy, providing a theoretical foundation for precision reconstruction. Based on quantitative indicators of neuronal reconstruction difficulty, a data-algorithm precise matching strategy was developed to adaptively select the most suitable reconstruction method for different types of neuronal data while leveraging the complementary strengths of multiple reconstruction algorithms. Experimental results demonstrated significant improvements in reconstruction accuracy across multiple data categories, with the best-performing image category achieving an accuracy improvement of up to 18.8%. Furthermore, a data-annotator precise allocation strategy was proposed to match data difficulty with annotator capability, enabling efficient human-machine collaborative reconstruction and transforming conventional experience-based reconstruction into a precision-driven quantitative reconstruction paradigm. Compared with traditional reconstruction strategies, the proposed allocation strategy improved reconstruction accuracy by 44.3% and increased overall reconstruction efficiency by 34.6%. In summary, the proposed framework enables quantitative evaluation and controllable assurance of neuronal reconstruction quality. By transforming neuronal reconstruction from a conventional single-method paradigm into a data-driven precision decision-making paradigm, the proposed approach substantially improves reconstruction efficiency while maintaining high reconstruction quality. This work provides reliable methodological support and a solid data foundation for large-scale neuronal morphology analysis and neural circuit research.

The ventral posteromedial thalamic nucleus (VPM) relays orofacial somatosensory signals to the primary somatosensory cortex (SSp), yet its functional heterogeneity remains poorly defined at single-cell resolution. We combined fluorescence micro-optical sectioning tomography (fMOST) with single-neuron whole-brain reconstruction to map projections of vesicular glutamate transporter 2-positive (VGluT2⁺) excitatory neurons in the ventrolateral subdivision of VPM (vlVPM). Using 88 high-quality neurons with validated completeness, we identified pronounced structural heterogeneity and classified two clusters with distinct axonal profiles. Notably, we report two key findings that challenge the classical view: vlVPM neurons preferentially target superficial SSp layers (L1-3) rather than layer 4 (L4), and a multi-target subpopulation simultaneously innervates the motor cortex and basal ganglia via collaterals, establishing direct anatomical links to motor regulatory networks. These anatomical findings demonstrate that vlVPM establishes diverse projection pathways beyond canonical sensory relays, enabling direct structural crosstalk between sensory and motor circuits. This work provides a single-cell resolution framework for interpreting the anatomical basis of thalamic involvement in sensorimotor coordination.

Electroencephalography (EEG) records electrical brain activity from the scalp and is widely used in brain-computer interface (BCI) systems for communication, and assistive technologies. EEG is widely used in motor-imagery (MI) based BCIs, where neural recordings contain highly individual and potentially sensitive information. In this regard, federated learning (FL) is a prominent privacy-enhancing approach which enables collaborative model training without centralising raw signals. However, recent work has shown that FL models still leak private information through membership inference attacks (MIAs). Most existing studies examine only single attack type, so it remains unclear how multiple MIAs together expose different layers of privacy risk in FL-based EEG systems. To address this gap, this study develops a federated MI-EEG classification framework and evaluates privacy leakage across four complementary MIAs: record-level, feature-level, gradient-level, and client-identity inference. Two neural networks were trained using per-subject FL, and differential privacy (DP) with epsilon (ε) ∈ {1, 5, 10} was applied to client updates. Results showed that standard FL alone provides limited intrinsic protection, while adding DP substantially reduces attack success particularly for gradient and identity-level attacks. Strong privacy settings (ε = 1) offered the greatest leakage reduction but degraded classification accuracy, whereas a moderate privacy budget (ε = 5) achieved the most favourable privacy-utility balance. Overall, the findings demonstrate that FL alone is insufficient as a privacy safeguard for EEG-BCI systems. Explicit privacy mechanisms such as DP are required to mitigate multi-level leakage, supporting the design of trustworthy and secure neural-learning technologies.

Explainable Artificial Intelligence (XAI) is gaining popularity in early diagnosis and monitoring of dementia. Herein, we recommend the incorporation of the National Institute of Mental Health's Research Domain Criteria (NIMH-RDoC) framework with XAI-informed diagnostic protocols to help establish diagnosis at early stages of Alzheimer's disease (AD). RDoC has a dimensional structure that extends across units of analysis from genes and molecules to circuits, physiology, behavior, and introspection. By restructuring diverse features as inputs including apolipoprotein E (APOE) genotype, amyloid and tau biomarkers, computational neuroimaging-informed cortical atrophy, Positron Emission Tomography (PET) hypometabolism, quantitative electroencephalography (qEEG) rhythms, cognitive tests, and digital behavioral markers), onto RDoC units provides more insightful and inclusive models. In this context, data-driven approaches such as XAI can achieve not only increased interpretability but also enhance their mechanistic validity. Such an innovative approach places data-driven model outputs within neurobiologically based domains such as Cognitive Systems, Negative Valence, and Arousal/Regulatory Systems. Our synthesis suggests that a 'converging RDoC and XAI' approach may help bolster the coherence of AD biomarkers, promote model exploration in clinical decision-making. This approach is also expected to provide a strategic roadmap for translational neuroscience and personalized medicine. Another major aim of this study is to critically analyze current XAI approaches used in dementia research, particularly the diagnostic and prognostic aspects. By explicitly grounding explanations in RDoC cognitive domains and paradigms, the framework also aims to make model outputs meaningful in terms of specific mental functions (e.g., episodic memory, cognitive control), thereby supporting neuropsychologically-informed diagnosis, categorization, and communication with patients and caregivers.

Deficits in neural information integration are hypothesized to underlie diverse psychiatric symptoms, yet the specific patterns of alteration across different disorders remain unclear. In this study, we decomposed information dynamics between brain regions into synergistic and redundant components using a recent information-theoretic approach based on Partial Information Decomposition and applied to resting-state fMRI data from individuals with schizophrenia (SZ), autism spectrum disorder (ASD), and attention-deficit/hyperactivity disorder (ADHD). Our analysis revealed distinct disorder-specific profiles: SZ and ASD exhibited a widespread reduction in synergy, whereas ADHD showed a contrasting increase. Furthermore, ASD was uniquely characterized by a significant reduction in redundancy. Meta-analytic functional annotation using NeuroSynth associated synergy with higher-order cognitive functions and redundancy with lower-level sensorimotor processing. To investigate multivariate organization of these patterns that distinguish psychiatric diagnoses, we employed Linear Discriminant Analysis (LDA). This analysis demonstrated that synergy and redundancy partially capture distinct dimensions of network variation, exhibiting substantial complementarity in their multivariate structure. While redundancy overlapped considerably with correlation-based connectivity, synergy reflected additional structure not fully represented by conventional measures. Together, these findings indicate that decomposing information dynamics provides complementary perspectives on large-scale network organization, offering a refined framework for characterizing psychiatric and neurodevelopmental disorders.

Many studies have shown that major depressive disorder (MDD) is highly relevant to the dysfunction of frontal lobes. Given the crucial role of brain network organization for MDD, this work focuses on inferring effective connectivity patterns and their changes in the frontal lobes of non-depressive and MDD subjects. We collected frontopolar left (Fp1), frontopolar right (Fp2), and frontal midline (Fz) channels of resting electroencephalogram (EEG) from 22 subjects with non-depressive states (non-depressive group), 23 subjects with mild depression (mild MDD group) and 12 subjects with moderate depression (moderate MDD group). Complete Physiology-Based Model was introduced to simulate the EEG signals of brain activities in the three targeted lobes. Then, spectral dynamic causal modelling was performed to investigate effective connectivity and its changes between the three targeted channels. The findings suggest that MDD groups exhibit a significant decline in the average strength of excitatory connectivity between pyramidal cells of the three lobes, compared to the non-depressed group. Additionally, the fluctuation of connectivity across trials among frontal lobes decreases in the mild and moderate MDD groups. The MDD groups also showed frontal asymmetry, specifically in the form of significant differences in bidirectional excitatory connections between Fp1 and Fp2. These effective connectivity changes of frontal lobes may indicate its clinical potential in future research on depression.

Electroencephalography (EEG) provides a non-invasive, portable, and cost-effective solution for emotion classification, but existing methods often struggle to capture spatial-spectral dependencies and multi-temporal scale dynamics inherent in EEG signals. This paper proposes a spectral-spatial attention multi-temporal scale network (SSA-MTSNet) to address these challenges. The SSA-MTSNet integrates 3 components: a spectral-spatial attention module, a multi-temporal scale spatio-temporal convolution module, and a long short-term memory (LSTM) module. First, the SSA-MTSNet preserves electrode topology and simultaneously enhances EEG signal frequencies and interactions among brain regions through attention mechanisms. Then, short- and long-term emotional cues are captured using multi-temporal scale convolution, followed by sequence modeling with LSTM. Evaluated on the SEED and SEED-IV datasets, the SSA-MTSNet achieves average accuracies of 98.34% and 91.79% respectively, and 95.13% and 95.30% on the valence and arousal classification tasks of DEAP, respectively. These results suggest that the proposed model achieves competitive performance across different experimental paradigms and emotion labeling strategies by jointly modeling complementary spectral, spatial, and temporal EEG information for emotion recognition.

Cross-subject motor imagery classification remains challenging due to EEG data scarcity and inter-subject variability. This study proposes a novel framework integrating generative data augmentation with domain adaptation. First, we employ a diffusion probabilistic model to generate high-fidelity synthetic EEG samples, effectively enriching the training data. Subsequently, we propose the AMSC-DANN architecture, which synergizes an Adaptive Multi-Scale Convolution (AMSC) module for extracting multi-granular features with a Domain Adversarial Neural Network (DANN). This combination enables the model to learn discriminative temporal-spectral representations while simultaneously aligning feature distributions across different subjects. Extensive experiments on BCI Competition IV datasets 2a and 2b demonstrate that our proposed framework outperforms state-of-the-art baselines, validating its effectiveness in enhancing cross-subject generalization.

Advances in brain imaging have generated unprecedented volumes of high-dimensional data, yet extracting meaningful information from complex, noisy, and incomplete brain imaging data remains a significant challenge. Diffusion models (DMs) have introduced a paradigm shift in this field, surpassing traditional generative approaches. This review systematically examines the theoretical foundations of diffusion models, and their practical applications in eight brain imaging computing tasks: registration, super-resolution, cross-modal reconstruction and synthesis, segmentation, classification, brain network analysis, brain-computer interface (BCI) signals augmentation, and BCI decoding. Additionally, we emphasize obstacles that hinder deployment in practice, including computational scalability and sampling inefficiency, limited generalization under domain shift sensitivity, as well as multimodal integration and alignment constraints, while outlining potential future directions that emphasize the convergence of diffusion models with large-scale foundation models, which hold the potential to advance scalable, reliable, and clinically embedded brain imaging solutions. Throughout this review, we aim to establish a roadmap of progress and translational hurdles to guide emerging research and accelerate collaboration spanning DMs, clinical brain imaging, and engineering disciplines.