搜索结果：brain-like

Three-dimensional (3D) brain-like models such as neurospheres and organoids recapitulate key aspects of human cortical development, providing valuable platforms for studying neurogenesis, disease mechanisms, and translational applications. However, their characterization has traditionally relied on two-dimensional (2D) histology, which fails to capture spatially clustered populations and complex cytoarchitecture. Tissue-clearing methods combined with advanced imaging now enable volumetric analyses that preserve the native 3D organization of neural cells. We applied the iDISCO+ clearing protocol to human iPSC-derived neurospheres and performed whole-mount immunolabeling followed by light-sheet fluorescence microscopy. Quantitative analyses of cell-type composition were benchmarked against conventional 2D cryosectioning. Temporal studies were performed from day 25 to day 60 of differentiation using layer-specific cortical markers (BRN2, CTIP2, and FOXP2) to assess the dynamics of superficial and deep cortical neuron generation. Finally, the impact of 3D analysis on marker co-localization was evaluated using spot-based spatial quantification of CTIP2 and COUP-TF1 expressions. Benchmarking volumetric quantification against conventional histology revealed that both methods produced comparable estimates for broadly distributed markers such as Ki67 and CTIP2. In contrast, 2D analysis substantially underestimated clustered populations, with SATB2+ and especially FOXP2+ neurons detected at significantly higher proportions in 3D. Building on this, temporal analysis of 3D-cleared neurospheres demonstrated a marked expansion of BRN2+ superficial neurons between day 25 and day 60, a robust late-stage increase in CTIP2+ deep-layer neurons, and a maturation-dependent rise of FOXP2+ layer VI neurons. In addition, 3D analysis enabled robust quantification of CTIP2+/COUP-TF1+ co-expressing neurons, revealing significantly higher estimates than 2D sectional analysis by integrating spatial information across the entire neurosphere volume. These dynamics underscore the capacity of 3D volumetric imaging to capture both homogeneous and spatially restricted neuronal populations, providing a faithful readout of cortical layer specification during neurosphere differentiation. Our study demonstrates that iDISCO+ combined with light-sheet microscopy provides a powerful and reliable approach for quantitative, multiscale characterization of neurospheres. By preserving spatial integrity and enabling volumetric assessment of both single-marker expression and co-localization, this approach reduces biases inherent to 2D sectioning and improves the accuracy of temporal and spatial analyses in 3D neural models. While limitations remain regarding antibody penetration and antigen sensitivity, volumetric approaches are essential for advancing the accuracy and reproducibility of 3D neural model analyses. The online version contains supplementary material available at 10.1038/s41598-026-41741-7.

Tauopathies, such as Alzheimer's disease and frontotemporal dementia, are common neurodegenerative diseases characterized by misfolding, hyperphosphorylation, and aggregation of tau. Molecular mechanisms underlying tauopathies are still poorly understood, which is in part due to a lack of human models autonomously developing major disease hallmarks. The formation of late-stage disease phenotypes may require adult tau isoform expression, which contributes to tau pathogenesis but is challenging to replicate in human stem cell-derived systems, thus impeding research on underlying mechanisms and drug development. Here, we show that induction of adult human brain-like 4R tau isoform expression enables cell-intrinsic formation of late-stage tauopathy hallmarks in induced pluripotent stem cell-derived neurons engineered to contain synergistic tau mutations without exogenous sources of tau pathology. Neurons accumulated seeding-competent and hyperphosphorylated tau in tangle-like structures. Furthermore, exclusive expression of mutant 4R in the absence of the 3R tau isoform disproportionately intensified pathology, resulting in abundant tau misfolding and aggregation. Last, we provide proof of principle that our model can be translationally applied both to test chemical disease modulators and evaluate human tau PET tracers. Collectively, our model corroborates the central role of 4R tau isoform expression for pathogenesis in human neurons and enables investigations to elucidate mechanisms underlying human tauopathy formation. Moreover, it may serve as a platform supporting urgently needed development of disease-modifying drugs.

The development of high-performance neuromorphic computing hardware is a key pathway to overcome the Von Neumann architecture's energy efficiency bottleneck. Among these hardware solutions, reconfigurable and low-power synaptic memtransistors are regarded as promising candidates for building high-energy-efficiency brain-like systems. Here, we fabricated an ultra-low-power, reconfigurable optoelectronic memtransistor using a vertical Nb-doped WSe2 (Nb-WSe2)/Te van der Waals (vdW) heterostructure. Under optical stimulation, the device can reproducibly emulate paired-pulse facilitation, short-term plasticity, and long-term plasticity, offering programmable AND/OR logic via combined photoelectronic control. Critically, the single-pulse energy consumption can decrease below 1 aJ, demonstrating excellent energy efficiency for optoelectronic synapses (about 4 orders of magnitude lower than that of the biological synapse). Utilizing wavelength-selective light pulses, we further emulate Pavlovian associative learning, highlighting the device's capability for multimodal synaptic conditioning. Finally, when used as the building block of a convolutional neural network (CNN), the memtransistor array achieves 92.32% accuracy on the CIFAR-10 benchmark. It retains 72.75% accuracy under multilevel noise, demonstrating strong classification performance and robustness. These results indicate that Nb-WSe2/Te vdW memtransistors are a promising candidate for constructing ultra-low-power, reconfigurable neuromorphic and brain-inspired computing hardware.

MedIntelliCare is an AI-powered medical assistant designed to enhance diagnostic accuracy, reduce cognitive load on healthcare professionals, and integrate real-time medical data. While current AI-driven medical systems focus on information retrieval and response generation, MedIntelliCare leverages Retrieval-Augmented Generation (RAG) combined with principles from neural computation and decision-making processes. This study explores the system's ability to simulate biologically inspired information processing by integrating brain-like computing, predictive modeling, and multimodal analysis, including EEG and neuroimaging data. By aligning MedIntelliCare with advances in computational neuroscience and intelligent diagnostics, we aim to establish a model that enhances clinical decision support through adaptive information retrieval. The system's future implications include cognitive disorder modeling, brain-computer collaboration, and advanced AI-driven diagnostics inspired by neural processing frameworks. Experimental validation using cosine similarity metrics demonstrates that MedIntelliCare achieves a 73% alignment with expert-generated reports, reinforcing its potential in neuro-inspired medical intelligence.

Emerging evidence implicates the oral-brain axis in neurodegeneration, yet large community-based studies remain limited. This study aimed to examine associations between periodontal health, oral microbiome, and cognitive performance, and to explore potential biological pathways underlying these relationships. We conducted a cross-sectional analysis of 1157 participants from the community-based Taizhou Imaging Study, all of whom underwent comprehensive periodontal examinations, salivary microbiome profiling, and cognitive assessments. Periodontal health and microbiome features were treated as exposures, and cognitive performance as the outcome. Associations between periodontal indices and cognitive scores were assessed using beta regression models adjusted for relevant confounders. Cognition-related microbial features were identified using Multivariate Associations with Linear Models (MaAsLin3), followed by mediation analyses to explore potential pathways linking periodontal health to cognitive function. Five clinical periodontal indices were found to be inversely associated with cognitive performance. Ten microbial genera (e.g., Haemophilus), 21 functional pathways (e.g., FoxO signalling), and two co-abundance modules, including a Treponema module, were significantly related to cognitive function. Mediation analysis suggested that 11 features, including nitrate-reducing taxa and a Treponema-driven inflammatory module, may partially mediate the relationship between periodontal health and cognition. These community-based findings reveal microbiome-mediated links along the oral-brain axis and highlight periodontal health and oral microbial homoeostasis as potential targets for early prevention of cognitive decline. This work was supported by the National Key R&D Program of China (2023YFC3606300), National Natural Science Foundation of China (82373658), Clinical Research General Project of the Shanghai Municipal Health Committee (202240355), Clinical Research General Project of Shanghai Municipal Health Commission (202440188), Noncommunicable Chronic Diseases-National Science and Technology Major Project (2023ZD0510000), Brain Science and Brain-like Intelligence Technology-National Science and Technology Major Project (2022ZD0211600).

Decoding human visual neural representations is scientifically important for advancing research on brain-like intelligence. Existing research typically aligns neural signals captured by fMRI or EEG with visual and linguistic features, thereby enabling models to decode brain activity into unseen visual categories. However, current methods still suffer from two fundamental challenges: (1) Representation drift. The alignment between learned features and new patterns degrades during continuous training, unlike the stable retention of knowledge in the human brain. (2) The incomplete modeling of common and individual representations. The model frequently fails to effectively disentangle common semantics from modality-specific information within image and text pairs. To address these issues, we propose a novel framework named MLHuB that mimics the human brain's learning mechanism. Firstly, we propose a memory unit responsible for reading and updating the learned text-image features as a way to consolidate acquired knowledge. Secondly, we compute common and individual features between text and images via orthogonal projection and utilize intra-modality mutual information maximization to regularize the learning of text-image pairs, encouraging the model to explore unseen knowledge. Finally, we integrate both intra- and inter-modality mutual information maximization to learn a more consistent joint representation between modalities. Extensive experiments on three datasets demonstrate that our proposed MLHuB achieves state-of-the-art performance.

According to in-depth research on the perception ability of dangerous omens of excellent drivers, references can be provided for the development of brain-like intelligence and its transplantation, as well as applications in the field of autonomous driving, which will improve the active safety and intelligence level of vehicles. Previous studies have shown that there is indeed a dangerous omen before an accident occurs. However, current studies are still unclear about the bio-psychophysiological characteristics exhibited by drivers with high levels of sensory agility when they anticipate potential warning signs, and there is no method for screening such drivers who can perceive dangerous omens proposed by any research. To address the above issues, this paper conducts in-depth research. Firstly, through designing dangerous scenarios and conducting hazard perception tests, we collect physiological, psychological, and physical data, such as drivers' bioelectrical signals (electroencephalogram and electrocardiogram) and eye movements. Secondly, through playing back experimental videos, actively questioning drivers, and analyzing local changes in their electroencephalogram data, the driver's ability to identify a dangerous omen and the moment of perception are determined. Thirdly, based on techniques such as the Kolmogorov-Smirnov test and the Mann-Whitney U test, the differences in bioelectrical and eye movement characteristics between drivers who can perceive a dangerous omen and others can be further revealed. Finally, the driver's bioelectrical and eye movement characteristics are used as latent variables, and their corresponding data are utilized as observation indicators. We construct a structural equation model for screening drivers capable of perceiving a dangerous omen and conduct calibration and validation. This study provides inspirational ideas for empowering vehicles to identify potential hazards, advancing end-to-end and other higher-level autonomous driving technologies, and further enhancing road traffic safety.

Understanding how the brain encodes stimuli has been a fundamental problem in computational neuroscience. Insights into this problem have led to the design and development of artificial neural networks that learn representations by incorporating brain-like learning abilities. Recently, learning representations by capturing similarity among input samples has been studied (Pehlevan et al., 2018) to tackle this problem. This approach, however, has thus far been used only to learn downstream features from an input and has not been studied in the context of a generative paradigm, where one can map the representations back to the input space, incorporating not only bottom-up interactions (stimuli→latent) but also learning features in a top-down manner (latent→stimuli). We investigate a kernel similarity matching framework for generative modeling. Starting with a modified sparse coding objective for learning representations proposed in prior work (Olshausen & Field, 1996; Tolooshams & Ba, 2021), we demonstrate that representation learning in this context is equivalent to maximizing similarity between the input kernel and a latent kernel. We show that an implicit generative model arises from learning the kernel structure in the latent space and show how the framework can be adapted to learn manifold structures, potentially providing insights as to how task representations can be encoded in the brain. To solve the objective, we propose a novel alternate direction method of multipliers (ADMM)-based algorithm and discuss the interpretation of the optimization process. Finally, we discuss how this representation learning problem can lead toward a biologically plausible architecture to learn the model parameters that ties together representation learning using similarity matching (a bottom-up approach) with predictive coding (a top-down approach).

Replicating brain-like computation with fluidic memristors offers advantages in energy efficiency and chemical responsiveness over solid-state devices, yet scaling remains challenging due to complex fabrication and their amorphous nature. Herein, we developed a confined hydrogel fluidic memristor by forming a gel-gel interface at the micropore orifice. This design with confined hydrogel enables scalable fabrication of a 10×10 fluidic memristor array (FMA) on polyimide micropores. FMA exhibits fundamental neuromorphic behaviors like paired-pulse facilitation/depression, spike-rate-dependent plasticity, and chemical-regulated plasticity. We also used reservoir computing algorithms with FMA to recognize both computer-generated black-and-white digit images and handwritten digits, achieving a classification accuracy of 89.5% on the Modified National Institute of Standards and Technology dataset. This study demonstrates a hydrogel confined fluidic memristor array, paving an avenue for creating large-scale fluidic memristor arrays and hardware intelligence with ions.

Most memristive circuits only consider single-channel memory, but dual-channel memory and temporal order memory are ignored. In this paper, a memristor-based circuit is proposed, which can realize dual-channel memory and temporal order memory. The designed circuit includes dual-channel memory module, temporal order memory module, feedback module and recollection module. Based on the three-level processing model of memory information, the processes of memory formation under single-channel and dual-channel are simulated separately. In addition, temporal order memory is simulated and the effect of time interval between stimuli on temporal order judgement is also taken into account. The impact of contextual source variability on temporal order memory is also considered. The simulation results in PSPICE confirm the circuit's ability to implement these bionic memory functions. This offers potential insights into the advancement of brain-like intelligence in fault detection systems. The circuit's potential application in industrial fault detection is certificated, where it can recognize patterns and detect anomalies in sensor data.

Objective.This study aims to design a CMOS-based circuit that mimics the behavior of real brain synapses, focusing on both plasticity and inhibition. The goal is to improve the biological realism and learning ability of neuromorphic hardware.Approach.A unified CMOS-based synaptic architecture is proposed that integrates short-term plasticity (STP) and long-term plasticity (LTP) with two forms of synaptic inhibition: divisive and subtractive. The STP circuit models short-term depression and facilitation, while the LTP mechanism employs spike-timing-dependent plasticity to capture temporally driven synaptic modifications. Furthermore, a spiking neuronal network is designed to demonstrate biologically accurate inhibitory effects and to perform max pooling via divisive inhibition. All circuits are implemented and simulated in TSMC 180 nm CMOS using Cadence Virtuoso.Main results.The proposed circuits successfully reproduce key biological features of synaptic behavior. The STP and LTP blocks enable time-dependent modulation of synaptic weights, while the inhibitory networks exhibit both divisive and subtractive control over postsynaptic firing frequency. The maxpooling operation, achieved via divisive inhibition, allows the target neuron to respond to the input with the highest spiking activity selectively. Simulation results confirm the correct functional behavior of all the designed circuits.Significance.This work provides a simple and effective hardware solution for modeling fundamental synaptic functions. It supports adaptive learning and efficient processing in neuromorphic systems. The results can help build better brain-like systems for AI, robotics, and brain-computer interfaces.

Neuromorphic computing, a highly promising computational architecture, has provided an efficient solution to overcome the limitations of storage-compute separation and scaling constraints. The key to implementing this architecture lies in the development of artificial neurons and synapses as core neuromorphic components capable of biomimicry. Diverse libraries of two-dimensional (2D) materials with atomic-scale thickness and rich tunable physicochemical properties have risen to prominence in recent years. These unique properties meet the critical requirements of neuromorphic devices for ultralow power consumption, dynamic plasticity, and multifunctional integration, thereby facilitating breakthroughs in next-generation high-performance and versatile neuromorphic hardware systems. In this paper, recent advances in dedicated artificial neuron and synapse devices based on 2D materials are reviewed, with a focus on biomimetic models, physical mechanisms, and performance metrics. The discussion further extends to sophisticated switching strategies in reconfigurable components. Then, the systemic integration of neuromorphic devices is summarized, with particular focus on their functional roles in neural perception, neural networks, and logical operation tasks. Finally, a systematic analysis of the limitations at the device and system levels for artificial neurons and synapses is presented, charting a roadmap toward more efficient and multifunctional brain-like chips.

Memory and forgetting in the human brain are multidimensional, synergistic, and dynamic processes involving neurophysiology and cognitive processes. To overcome the limitations of static and non-stimulation responsiveness of traditional dry and rigid information storage media, which are static and unresponsive to external stimuli, hydrogels with wet and soft characteristics with brain tissue have become one of the research hotspots in recent years for applications in information storage and dynamic memory materials. Inspired by the processes of memory forgetting and relearning in the human brain in cognitive psychology, we propose a structurally simple dopamine-chitosan memory film with forgetting capability. This memory film was fabricated through a two-step electrodeposition and electrooxidation process, simulating the brain's slow forgetting (long-term memory) and relearning mechanisms based on the gradual oxidation and reducibility of dopamine over time. The information storage of memory films was quantified by detecting response currents via electrochemical methods, successfully reproducing key features of the Ebbinghaus forgetting curve and the spaced repetition learning curve (antagonistic forgetting curve). The results show that the memory film maintains good long-term memory retention and repetitive learning functionality, potentially inspiring future studies on brain-like memory forgetting using wet, soft materials and redox-active compounds.

Mixed protonic-electronic conductors (MPECs) have been developed to maximize static conductivity for electrochemical applications, but emerging applications that leverage proton-electron coupling (PEC) require dynamic conductivity control. To achieve this, we propose a "de-doping" strategy in a hydrogen-bonded coordination polymer {[Co(DMF)2(H2O)2(bipy)](NO3)2·2(DMF)}n (bipy = 4,4'-bipyridine, DMF = N,N-dimethylformamide) named Co-BAND. By isostructural substitution of Ni(II) (d8) in the established Ni-BAND with Co(II) (d7), we designed Co-BAND to suppress the intrinsic conductivity while preserving proton transport and PEC. As a result, Co-BAND exhibits a giant conductivity modulation (1.15 × 106) in response to humidity changes and implements complex brain-like learning rules. We also demonstrate chemical control of synaptic plasticity via solvent vapor exposure. This biomimetic neuromodulation tunes transport and learning rules based on vapor polarity, proticity, and steric effects. This work establishes conductivity modulation as an important design metric for MPECs and highlights their potential as designable platforms for stimuli-responsive applications.

A structured-light robotic system enables noninvasive surface-based facial registration for patient-to-image alignment; however, it has not been widely adopted for basal ganglia hematoma treatment. This study evaluates its safety, accuracy, and efficiency in patients with basal ganglia hematomas. Retrospective consecutive patients with spontaneous basal ganglia hematomas (10-30 mL) admitted from January 2022 to January 2024 were grouped as surgical or non-surgical. Outcomes were analyzed, focusing on registration accuracy and surgical efficiency. Baseline preoperative hematoma volume, GCS, NIHSS, and coagulation disorders were comparable between groups (p&#x2009;>&#x2009;0.05). Mean registration time was 120&#x2009;&#xb1;&#x2009;56&#xa0;s and surgery time 102.22&#x2009;&#xb1;&#x2009;6.33&#xa0;min; hematoma clearance was 72% initially and 93% at 7 days. Entry-point discrepancy was 2.2&#x2009;&#xb1;&#x2009;0.7&#xa0;mm; catheter tip alignment in the upper/middle/lower hematoma segments was 4%/80%/16%, with a maximum 5.7-mm deviation from the midline axis. The surgery group showed faster recovery with higher 24&#xa0;h-GCS (p&#x2009;<&#x2009;0.001) and better NIHSS/ADL/mRS than non-surgery (p&#x2009;<&#x2009;0.001 at all time points; p&#x2009;=&#x2009;0.003 and p&#x2009;=&#x2009;0.026 at 30 and 90 days; p&#x2009;=&#x2009;0.041). At 90 days, facial palsy, sensory loss, and ulcers were similar (p&#x2009;>&#x2009;0.05), whereas pneumonia (p&#x2009;=&#x2009;0.034) and UTIs (p&#x2009;=&#x2009;0.002) were lower in the surgery group. No mortality, bleeding, or reoperations occurred. Surface registration using the 3D structured light technique is a fast and precise alternative treatment for selected patients with spontaneous medium-volume basal ganglia hematomas, which can improve clinical efficiency while maintaining sufficient accuracy and safety to meet clinical requirements. retrospectively registered with approval by the Ethics Committee of Shanghai Fourth people's hospital, Tongji University School of Medicine (No. 2022327-001).

A growing body of research has recognized the heterogeneity among aggression perpetrators, particularly regarding their social status and victimization experiences. However, little is known about how distinct perpetrator roles evolve over time or the ecological factors that shape their heterogeneity and developmental dynamics. To address these gaps, the classification and transition patterns of aggression perpetrators in a sample of 2578 Chinese adolescents (Mage&#x2009;=&#x2009;12.99, SD&#x2009;=&#x2009;0.60; 47.9% girls) were examined in this study over a 2-year period. A latent class analysis revealed three perpetrator subgroups: high-status aggressors (7.22%), low-status aggressors (12.45%), and low-status aggressive victims (9.15%). Furthermore, high-status aggressors showed short-term stability but tended to shift to low-status roles in the long term, whereas low-status aggressors and aggressive victims exhibited instability across both time intervals. Analyses of multisystemic predictors revealed that protective and risk factors exerted differential effects across perpetrator subgroups, shaping both the persistence and developmental transitions of certain roles. These findings emphasize the need for subgroup-specific intervention strategies that integrate resources systematically at the individual, familial, school, and community levels.

暂无摘要（点击查看详情）

The role of brain network dynamics in relation to amyloid beta (A&#x3b2;) and tau pathology across Braak stages remains unclear. In this cross-sectional study of 216 participants from Translational Biomarkers of Aging and Dementia (TRIAD) cohort, we analyzed resting-state functional magnetic resonance imaging using a multilayer modularity algorithm to assess brain network dynamics across 10 predefined functional networks, stratified by amyloid and tau positron emission tomography biomarkers and Braak stages. Switching rates were significantly elevated in A&#x3b2;-positive/tau-positive individuals relative to A&#x3b2;-negative/tau-negative individuals, and increased progressively with advancing Braak stages. Elevated switching rates were strongly correlated with A&#x3b2; and tau burden in dorsal attention network and sensorimotor network, as well as with cognitive severity. Importantly, the interaction between network switching rate and A&#x3b2; burden synergistically contributed to accelerated tau accumulation in Braak stage III to V regions. These findings support the framework that increased network switching may amplify A&#x3b2;-related tau load and cognitive deterioration in Alzheimer's disease.