Brain-X最新文献

This comprehensive review aims to clarify the growing impact of Transformer-based models in the fields of neuroscience, neurology, and psychiatry. Originally developed as a solution for analyzing sequential data, the Transformer architecture has evolved to effectively capture complex spatiotemporal relationships and long-range dependencies that are common in biomedical data. Its adaptability and effectiveness in deciphering intricate patterns within medical studies have established it as a key tool in advancing our understanding of neural functions and disorders, representing a significant departure from traditional computational methods. The review begins by introducing the structure and principles of Transformer architectures. It then explores their applicability, ranging from disease diagnosis and prognosis to the evaluation of cognitive processes and neural decoding. The specific design modifications tailored for these applications and their subsequent impact on performance are also discussed. We conclude by providing a comprehensive assessment of recent advancements, prevailing challenges, and future directions, highlighting the shift in neuroscientific research and clinical practice towards an artificial intelligence-centric paradigm, particularly given the prominence of Transformer architecture in the most successful large pre-trained models. This review serves as an informative reference for researchers, clinicians, and professionals who are interested in understanding and harnessing the transformative potential of Transformer-based models in neuroscience, neurology, and psychiatry.

Caused by severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), coronavirus disease 2019 (COVID-19) primarily manifests as respiratory dysfunction. However, emerging evidence suggests SARS-CoV-2 can invade the brain, leading to cognitive impairment (CI). It may spread to other brain regions through transsynaptic neurons, including the olfactory, optic, and vagus nerves. Moreover, it may invade the central nervous system through blood transmission or the lymphatic system. This review summarizes the neuroimaging evidence from clinical and imaging studies of COVID-19-associated CIs, including magnetic resonance imaging and 18F-fluorodeoxyglucose positron emission tomography-computed tomography. The mechanisms underlying COVID-19-associated CIs are currently being actively investigated. They include nonimmune effects, such as viral proteins, tissue hypoxia, hypercoagulability, and pathological changes in neuronal cells, and immune effects, such as microglia and astrocyte activation, peripheral immune cell infiltration, blood-brain barrier impairment, cytokine network dysregulation, and intestinal microbiota. Inflammation is the central feature. Both central and systemic inflammation may cause acute and persistent neurological changes, and existing evidence indicates that inflammation underlies the elevated risk of Alzheimer's disease. Finally, potential therapeutic options for COVID-19-associated CIs are discussed. In-depth research into the pathological mechanisms is still needed to help develop new therapies.

This study aimed to evaluate [¹⁸F]GSK1482160 Positron emission tomography imaging for targeting P2X7R, a biomarker for neuroinflammation. Studies of acute neuroinflammation in rodents and transgenic mice with Alzheimer's disease (AD), as well as wild-type (WT) controls, were conducted via PET-CT-MRI scans after tail vein injection of [¹⁸F]GSK1482160. Imaging was quantified based on the time-activity curve, the standardized uptake value ratio, and the binding kinetics distribution volume ratio (DVR) to assess the expression of P2X7R. Tissues were collected post-PET for immunofluorescence staining. Correlation analysis was performed between DVR and Morris water maze test results. Finally, dynamic Positron Emission Tomography-Magnetic Resonance Imaging (PET-MRI) scans were performed in healthy non-human primates (NHPs). Our study demonstrated that AD mice had a significantly higher DVR than WT mice in the hippocampus (0.92 ± 0.06 vs. 0.79 ± 0.02, p < 0.05), cortex (1.09 ± 0.03 vs. 0.88 ± 0.04, p < 0.05), and striatum (1.02 ± 0.10 vs. 0.83 ± 0.1, p < 0.05). Immunofluorescence staining showed increased expression of P2X7R in the AD, along with its colocalization with activated microglia and astrocytes. Correlation analysis indicated that brain regions with higher binding of [¹⁸F]GSK1482160 (i.e., the cortex, striatum, and hippocampus) were more vulnerable to cognitive impairment. PET-MRI scans of healthy NHPs demonstrated the feasibility of brain penetration and P2X7R target engagement for the translation of [¹⁸F]GSK1482160 in human studies.

Hybrid neurophysiological signals, such as the combination of electroencephalography (EEG) and electromyography (EMG), can be used to reduce road traffic accidents by obtaining the driver's intentions in advance and accordingly applying appropriate auxiliary controls. However, whether they can be used in combination and can achieve better results in situations of detecting emergency braking from normal driving and soft braking has not been explored. This study used one feature-level (hybrid BCI-FL) and three classifier-level (hybrid BCIs-CLs) hybrid strategies, the spectral band, and spectral point features to construct recognition models. Offline and pseudo-online experiments were conducted. The recognition performance with the spectral point features showed a better result than that with spectral band features. In all experiments, the two proposed hybrid BCI strategies could achieve a detection accuracy close to or above 95%, while the detection advanced time is less than 300 ms. In particular, for the developed hybrid BCI recognition models, the hybrid BCI-FL and hybrid BCI-CL2 recognition models with spectral point features achieved 4.25% (p < 0.015) and 4.69% (p < 0.006) higher system accuracies, respectively, than that of the current better single EMG-based recognition model. This research promotes the application of hybrid EEG and EMG signals in intelligent driving assistance systems.

Illustration of brain-inspired AI-driven scientific research: predicting new information, discovering novel therapies, and designing new materials.

Network analysis, an interdisciplinary method rooted in graph theory and complex systems, is a promising approach for advancing our understanding of the brain's complex architecture and its implications for behavior, cognition, and mental health. Network analysis transcends the traditional psychiatric diagnostic model, which oversimplifies mental disorders by treating them as distinct physical illnesses, often creating an “epistemic prison” that fails to account for the nuanced interplay between neurological, biological, psychosocial, and cultural influences shaped by patients' life experiences.¹ By mapping and examining the intricate network of neuronal connections and larger brain region interactions, network analysis offers deep insights into brain communication pathways, their role in cognitive function, and how their disruption may lead to neurological disorders. Despite the potential of this method, the application of network analysis in brain science is underutilized, highlighting the need for increased awareness and the development of network-based studies to fully realize its transformative potential for behavior and brain research. Therefore, we introduce an insightful behavioral exemplar to increase awareness of the potential application of network analysis in brain science.

In their landmark study, Hu et al. not only challenged the compartmentalization of psychiatric diagnoses but also provided a novel lens through which we can view mental disorders from a neurobiological perspective.² By employing network analysis, they illustrated that psychiatric symptoms occur in isolation but as a part of a complex network at the behavioral level, significantly resonating with a variety of human brain functions and structures. This approach underscores the centrality of the motivation and pleasure factor, which is potentially linked to the brain's reward system, and its significant impact on broader cognitive and social functioning across different psychiatric conditions. The study integrated the transdiagnostic model with sophisticated statistical methods, such as the least absolute shrinkage and selection operator, further elucidating ways to examine potential intricate brain–behavior relationships in the future.³ Such neuroscientific insights pave the way for a more nuanced understanding of psychopathology; additionally, they can inform targeted interventions that can modulate specific neural circuits implicated in multiple psychiatric disorders.

Although network analysis was employed behaviorally in this study, it offers methodological breakthroughs for prospective neurological studies, allowing for a unified representation of complex brain functions and statistically significant control over variables of interest. It illuminates how alterations in one node can reverberate throughout the entire network, providing a level of insight traditional models have f

Neurodegenerative diseases (NDs) stand for a group of disorders characterized by the progressive loss of neurons in the brain and peripheral organs, resulting in motor and cognitive dysfunction. The global prevalence of NDs, including Alzheimer's disease, Parkinson's disease, Huntington's disease, and amyotrophic lateral sclerosis, is on the rise globally, primarily due to an aging population, positioning NDs as an increasing significant public health concern. Despite intensive research, few effective therapies that prevent or delay ND progression have been developed. Mounting evidence indicates that one of the well-defined risk factors for NDs is type 2 diabetes mellitus, and insulin resistance has also been proven to be related to cognitive decline. Certain antidiabetic drugs, such as glucagon-like peptide-1 receptor agonists, peroxisome proliferator-activated receptor gamma agonists, and metformin, have shown promise in offering neuroprotective benefits and alleviating ND symptoms beyond their glucose-lowering effects. Although the exact mechanisms remain elusive, these drugs offer a promising novel strategy for managing cognitive disorders. In this review, we first highlight the benefits and specific neuroprotective effects of multiple antidiabetic drugs and discuss the main mechanisms of action of antidiabetic drugs in treating NDs. These mechanisms include reducing protein aggregation and improving apoptosis, mitochondrial dysfunction, oxidative stress, and neuroinflammation. Finally, we summarize clinical trials evaluating these drugs for treating NDs.

Alzheimer's disease (AD), the most common form of dementia, is a progressive neurodegenerative disease characterized by cognitive deficits, β-amyloid (Aβ) accumulation-induced amyloid plaques, and tau hyperphosphorylation-induced neurofibrillary tangles.¹ Interestingly, emerging evidence suggests other factors may contribute to Aβ-associated pathologies.² β2-microglobulin (β2M), one of the major histocompatibility complex class I molecules, is a short peptide with seven antiparallel β-strands. It is elevated in AD brains and has recently been detected in the amyloid plaque core.³ Therefore, increasing evidence suggests β2M may be a potential factor that promotes Aβ aggregation and neurotoxicity.

A recent study in Nature Neuroscience conducted by Zhao et al. found that β2M may be a possible factor involved in amyloid pathologies.³ The authors characterized the pathological changes of β2M and elucidated the functional involvement of β2M in amyloid deposition and spreading and in boosting Aβ neurotoxicity.³ They concluded that β2M is an essential coaggregation factor with Aβ in amyloid pathology and β2M-Aβ coaggregation is a therapeutic target for AD. In addition, their findings indirectly support the amyloid hypothesis and provide additional information underlying Aβ aggregation and Aβ neurotoxicity. In the past 2 decades, all clinical trials based on the amyloid hypothesis on AD have failed, prompting reconsideration of the amyloid hypothesis.³ However, the current study performed by Zhao et al. confirmed that inhibition of Aβ deposition significantly improves cognitive function, indirectly supporting this hypothesis. More importantly, their findings revealed that β2M expressed in the central nervous system and peripheral tissues are potential targets for alleviating amyloid pathology and Aβ neurotoxicity. Disrupting the β2M-Aβ interactions ameliorated Aβ deposition and Aβ-associated pathogenesis, exhibiting a tremendous therapeutic potential for AD treatment. Overall, although Zhao et al. cannot exclude the possibility that MHC class I contributes to β2M-dependent neurotoxicity, their study identifies a previously undefined role of β2M in Aβ aggregation and neurotoxicity and offers a novel therapeutic strategy for AD by inhibiting peripheral β2M (Figure 1).

Meanwhile, their findings also raise several intriguing questions that deserve further investigation. First, Zhao et al. discovered β2M is mainly present in microglia, suggesting it would be interesting to study further the relationship between microglial β2M and microglial function. For example, it is of great interest to investigate the role of β2M in microglial-mediated phagocytosis, synapse pruning, and neuroinflammatory response in AD and other brain disorders. Moreover, single-cell technologies have found various phenotypes

Advanced technologies that can establish intimate, long-lived functional interfaces with neural systems have attracted increasing interest due to their wide-ranging applications in neuroscience, bioelectronic medicine, and the associated treatment of neurodegenerative diseases. A critical challenge of significance remains in the development of electronic platforms that offer conformal contact with soft brain tissue for the sensing or stimulation of brain activities and chronically stable operation in vivo, at scales that range from cellular-level resolution to macroscopic areas. This review summarizes recent advances in this field, with an emphasis on the use of demonstrated concepts, constituent materials, engineered designs, and system integration to address the current challenges. The article begins with an overview of recent bioelectronic platforms with unique form factors, ranging from filamentary probes to conformal sheets and three-dimensional frameworks for alleviating the mechanical mismatch between interface materials and neural tissues. Next, active interfaces which utilize inorganic/organic semiconductor-enabled devices are reviewed, highlighting various working principles of recording mechanisms including capacitively and conductively coupled sensing enabled by high transistor matrices at high spatiotemporal resolution. The subsequent section presents approaches to biological integration which use active materials for multiplexed addressing, local amplification and multimodal operation with high-channel-count and large-scale electronic systems in a safe fashion that provides multi-decade stable performance in both animal models and human subjects. The advances summarized in this review will guide the future direction of this technology and provide a basis for next-generation chronic neural interfaces with long-lived high-performance operation.

Neuromorphic computing, benefitting from its integration of computing with memory, enables highly efficient parallel-computing capabilities. While artificial intelligence chips are expensive due to their large area and power consumption, neuromorphic devices have shown energy efficiency and compatibility with complementary metal-oxide-semiconductor transistor technology.¹ Complex neuronal circuits with feedforward and feedback topologies are the foundation for nonlinear information integration and processing in the human brain. In addition, the nonlinear integration of neuronal signals as the basic functions of the human brain's nervous system is also essential to implement machine learning. However, artificial neurons still face the challenge of nonlinearly integrating feedforward and feedback signals. It is crucial to develop bio-plausible neurons capable of those functions, including nonlinearity and integration of excitatory and inhibitory postsynaptic signals. Writing in Nature Nanotechnology, G. S. Syed and coworkers recently reported a major step toward bio-plausible optomemristive feedback neurons, enabling the simultaneous existence of separate feedforward and feedback paths within a neural network.²

The authors designed a delicate capacitor-like device with a 2D vertical heterostructure in which WS₂/MoS₂ and graphene served as the neuronal membrane and soma (Figure 1B), respectively. Generally, trapped electrons and holes in the WS₂/MoS₂ heterostructure recombine upon a positive back gate voltage (Figure 1A). The conductance state of p-doped graphene would further increase, representing an excitatory operation. In this work, the electron-hole carriers in the WS₂/MoS₂ heterostructure are easily separated upon illumination (Figure 1C), and the electrons are injected into graphene. The Fermi-level movement toward the Dirac point decreases the conductance of graphene, having an inhibitory effect. Specifically, graphene's gradual conductance change can be separately modulated through electrical and optical means (Figure 1D), mimicking excitatory and inhibitory functionalities. 2D memristors have been investigated to emulate leaky-integrate-and-fire feedforward neurons.³ The synergistic effect of both input signals mimics a competitive neuron and enables the simultaneous existence of separate feedforward and feedback paths within the neural network.

The winner-take-all (WTA) neural network is a critical computational model for artificial neural networks, which can be used to implement unsupervised competitive learning and cooperative learning. The traditional memristors make it difficult to separately process feedforward and feedback neuronal signals, necessitating peripheral circuits or software to mimic inhibition behavior. The developed optomemristive feedback neuron can respond to both el