Deformable image registration is a critical technology in medical image analysis, with broad applications in clinical practice such as disease diagnosis, multi-modal fusion, and surgical navigation. Traditional methods often rely on iterative optimization, which is computationally intensive and lacks generalizability. Recent advances in deep learning have introduced attention-based mechanisms that improve feature alignment, yet accurately registering regions with high anatomical variability remains challenging. In this study, we proposed a novel unsupervised deformable image registration framework, LGANet++, which employs a novel local-global attention mechanism integrated with a unique technique for feature interaction and fusion to enhance registration accuracy, robustness, and generalizability. We evaluated our approach using five publicly available datasets, representing three distinct registration scenarios: cross-patient, cross-time, and cross-modal CT-MR registration. The results demonstrated that our approach consistently outperforms several state-of-the-art registration methods, improving registration accuracy by 1.39% in cross-patient registration, 0.71% in cross-time registration, and 6.12% in cross-modal CT-MR registration tasks. These results underscore the potential of LGANet++ to support clinical workflows requiring reliable and efficient image registration. The source code is available at https://github.com/huangzyong/LGANet-Registration.

Aims Temporal expectations are considered as implicit timing, which is different from explicit timing. Furthermore, temporal expectations could be divided into exogenous and endogenous temporal expectations. However, it is still unclear about the neural activation under temporal expectations.Methods In the present study, an experimental paradigm was designed for eliciting the related brain activation under exogenous temporal expectations. Three conditions were used for the exogenous temporal expectations task. In order to compare the exogenous temporal expectations related activations to the endogenous', a proper endogenous temporal expectations task was used. Brain activations were obtained by using functional magnetic resonance imaging (fMRI).Results Exogenous temporal perception-related regions, including TPJ, MTG, thalamus, IFG, caudate, cuneus, SOG, calcarine, FEF, and SPL have a good agreement with previous studies. Furthermore, it shows that the precuneus, PCC, Brodmann area 8 (BA8), ACC, and BA10 were also activated, which overlap with regions of the mesial of the so-called "default mode network". Negative correlated activations to exogenous temporal expectations task (use an endogenous temporal expectations task as an analysis baseline) were also evaluated.Conclusion We found that the sum of exogenous and endogenous temporal expectations related cerebral regions was almost the same when compared to resting-state networks (RSNs). We propose that the cerebrum could activate in two modes for cognition: one is based on endogenous temporal expectations, and another is based on exogenous temporal expectations.

Personalized functional imaging using fMRI is a paradigm shift in neuroscience but demands extensive data per participant. Functional near-infrared spectroscopy (fNIRS) is a cost-effective alternative, yet its ability to capture individual-specific functional architecture with adequate precision remains unproven. To assess fNIRS for personalized mapping, we charted functional organization in nine individuals using task-based and resting-state fNIRS alongside fMRI. Each participant contributed 11.5 hours of imaging data, including 4 hours of resting-state and 4.2 hours of task-based (motor, working memory, language) fNIRS data, and 3.3 hours of matched fMRI for cross-validation. Results show fNIRS reliably identifies individual-specific resting-state networks, requiring only 20 minutes of data for stable characterization. Additionally, fNIRS captures subject-specfic activation patterns in sensorimotor and higher-order domains, including atypical right-lateralized language dominance. These personalized fNIRS profiles align strongly with fMRI, offering comparable reliability at far lower cost. fNIRS proves a reliable, scalable, and cost-efficient tool for personalized brain mapping.

Multi-modal Medical Image Fusion (MMIF) is a visual task that integrates multiple images from various imaging devices into a unified image, and its development is greatly limited by distorted pixel-level registration. Most existing generative approaches are built on encoder-decoder representation architectures, which can result in mismatches in structure and dimensions among modalities. In this paper, we propose StrFilter, a novel multi-modal medical image fusion framework that learns latent representations with complementary structures from heterogeneous images. Our key insight is to perceive complementary and shared features into a fusion sequence, and model the differences oriented by the distance structure between aligned feature spaces and data spaces. Specifically, StrFilter learns the necessary information to aggregate multidimensional embeddings via structured state-space mamba, a sequence modeling framework that adaptively filters redundant feature information and enhances the perception of organs and their boundaries in medical images. StrFilter then learns the Gromov-Wasserstein distance, a metric to align distributional differences in different spaces, constrained by a loss of a distance structure and a minimum handling cost. In this way, the generative model can benefit from multi-modal image prior information and meaningful structures. Experiments show that StrFilter effectively provides clearer boundaries and greater visual consistency for fused images.

Background: Deep learning has demonstrated significant potential for automated brain metastases (BM) segmentation; however, models trained at a singular institution often exhibit suboptimal performance at various sites due to disparities in scanner hardware, imaging protocols, and patient demographics. The goal of this work is to create a domain adaptation framework that will allow for BM segmentation to be used across multiple institutions. Methods: We propose a VAE-MMD preprocessing pipeline that combines variational autoencoders (VAE) with maximum mean discrepancy (MMD) loss, incorporating skip connections and self-attention mechanisms alongside nnU-Net segmentation. The method was tested on 740 patients from four public databases: Stanford, UCSF, UCLM, and PKG, evaluated by domain classifier's accuracy, sensitivity, precision, F1/F2 scores, surface Dice (sDice), and 95th percentile Hausdorff distance (HD95). Results: VAE-MMD reduced domain classifier accuracy from 0.91 to 0.50, indicating successful feature alignment across institutions. Reconstructed volumes attained a PSNR greater than 36 dB, maintaining anatomical accuracy. The combined method raised the mean F1 by 11.1 Conclusions: VAE-MMD effectively diminishes cross-institutional data heterogeneity and enhances BM segmentation generalization across volumetric, detection, and boundary-level metrics without necessitating target-domain labels, thereby overcoming a significant obstacle to the clinical implementation of AI-assisted segmentation.

Despite huge investment, therapies for brain disorders remain largely ineffective in clinical practice. Accumulating evidence indicates that this low translational success is closely linked to the long-standing overlook of the brain extracellular space (ECS) in preclinical research, clinical practice, and regulatory frameworks. After over 4 decades of scientific exploration, particularly with recent breakthroughs in imaging and quantitative measurement methods, it is timely to integrate the ECS into the current neuroscience framework. This paper investigates underlying determinants of low translational success of central nervous system drugs and therapeutic devices, reviews the historical and technical bottlenecks that lead to the neglect of ECS research, and emphasizes its transformative potential in reshaping therapeutic strategies. We propose incorporating the ECS into neuroscience research, clinical regulatory assessment, and medical education, thereby establishing a comprehensive paradigm that omits no physical space for precision therapeutics targeting brain disorders.

Inferring kinetic parameters in dynamic positron emission tomography (PET) imaging is challenging due to the significant noise, the complexity of compartment models, the high spatio-temporal resolution of the data, and the variability in framing protocols. This study proposes a robust neural network-based method that directly solves the ordinary differential equations (ODEs) derived from kinetic models. By discretizing the derivative term using finite difference methods, ODEs are parameterized as a four-layer neural network. The network is optimized by minimizing a specialized loss function that incorporates physiological constraints. The delay correction and parallel network processing are also incorporated for voxel-level parametric PET imaging. Both simulations and real patient studies confirm the method’s effectiveness and practical applicability. Compared with the reference methods, our method excels in accuracy, robustness and physiological interpretability, particularly in high-noise conditions. Overall, our proposed approach offers enhanced diagnostic capabilities and a deeper understanding of physiological processes, showing great potential for medical imaging and personalized treatment strategies. The code has been made publicly available at https://github.com/PKU-MIPET/NNAKPI.git.

Abnormal cognitive aging is characterized by memory decline beyond normal age-related physiological changes. Nevertheless, the intrinsic mechanisms driving individual memory heterogeneity during aging remain poorly elucidated. Given the critical roles of brain interstitial fluid (ISF) dynamics and extracellular space (ECS) transport in maintaining neural homeostasis, the present study aimed to explore whether ECS compartmentalization and ISF drainage disturbance contribute to interindividual variations in memory performance during aging. Aged rats were stratified according to their memory performance. A multi-modal strategy combining behavioral assessment, microdialysis, metabolomics, and electrophysiological recording was applied to detect structural and functional deficits in ECS barrier integrity and ISF drainage, with the assistance of MRI-based tracer imaging, fluorescence imaging, and ultrastructural observation. The results demonstrated that memory-impaired aged rats exhibited region-specific neurotransmitter imbalance in the caudate nucleus and thalamus, which was closely associated with age-related ECS barrier dysfunction. Such barrier dysfunction induced ISF drainage disturbance, thereby reducing local neurotransmitter concentrations in the caudate nucleus and impairing thalamocortical oscillations during non-rapid eye movement sleep, and ultimately disturbing memory consolidation. Further tracer imaging and electron microscopy examinations confirmed that compromised myelin integrity in the internal capsule served as the structural basis for age-related ISF drainage disturbance. Collectively, these findings reveal that ISF drainage disturbance induced by age-related ECS barrier alterations acts as a non-degenerative mechanism underlying memory heterogeneity in cognitive aging. This work highlights ECS compartment integrity as a promising biomarker and therapeutic target for intervening age-related cognitive decline.

Functional magnetic resonance imaging (fMRI) is crucial for studying brain function and diagnosing neurological disorders. However, existing analysis methods suffer from reproducibility and transferability challenges due to complex preprocessing pipelines and task-specific model designs. Here we introduce the Neuroimaging Foundation Model with Spatial-Temporal Optimized and Representation Modelling (NeuroSTORM), which learns generalizable representations directly from four-dimensional fMRI volumes and enables efficient transfer to diverse downstream applications. Specifically, NeuroSTORM is pretrained on 28.65 million fMRI frames from over 50,000 participants, spanning multiple centres and ages 5-100. It combines an efficient spatiotemporal modelling design and lightweight task adaptation to enable scalable pretraining and fast transfer to downstream applications. We show that NeuroSTORM consistently outperforms existing methods across five downstream tasks, including demographic prediction, phenotype prediction, disease diagnosis, re-identification and state classification. On two multihospital clinical cohorts with 17 diagnoses, NeuroSTORM achieves the best diagnosis performance while remaining predictive of psychological and cognitive phenotypes. These results suggest that NeuroSTORM could become a standardized foundation model for reproducible and transferable fMRI analysis.

Cue-induced craving is a core driver of addiction and relapse, and its significant heterogeneity represents a major barrier to precision intervention. Currently, there remains a lack of objective, quantifiable, and individualized neurobiological biomarkers. Here, we employed task-based electroencephalography (EEG) to capture the dynamic neural signatures underlying cue-induced craving in patients with heroin use disorder (HUD) and developed an individualized functional connectivity (FC)-based prediction model. We identified β-band power envelope connectivity (PEC) as a reliable biomarker capable of estimating subjective craving severity at the individual level. Notably, even after FC reconfiguration induced by intermittent theta burst stimulation (iTBS) over the left dorsolateral prefrontal cortex (L-DLPFC) or precuneus (PCu), the PEC-based framework's prediction of immediate craving levels following these perturbed states remained effective. Crucially, baseline β-band PEC demonstrated strong prognostic value for improvements in craving scores (L-DLPFC-iTBS: r = 0.856, P < 0.001; PCu-iTBS: r = 0.675, P = 0.008). This individualized predictive model was further validated in an independent dot-probe task dataset, demonstrating its generalizability across distinct cue-induced craving paradigms. Together, our study demonstrates that EEG FC features predict individual cue-induced craving levels and intervention outcomes, facilitating the advancement of digital biomarker-driven precision medicine.

Computed tomography (CT) is a cornerstone imaging modality for non-invasive, high-resolution visualization of internal anatomical structures. However, when the scanned object exceeds the scanner's field of view (FOV), projection data are truncated, resulting in incomplete reconstructions and pronounced artifacts near FOV boundaries. Conventional reconstruction algorithms struggle to recover accurate anatomy from such data, limiting clinical reliability. Deep learning approaches have been explored for FOV extension, with diffusion generative models representing the latest advances in image synthesis. Yet, conventional diffusion models are computationally demanding and slow at inference due to their iterative sampling process. To address these limitations, we propose an efficient CT FOV extension framework based on the image-to-image Schrödinger Bridge (I^2SB) diffusion model. Unlike traditional diffusion models that synthesize images from pure Gaussian noise, I^2SB learns a direct stochastic mapping between paired limited-FOV and extended-FOV images. This direct correspondence yields a more interpretable and traceable generative process, enhancing anatomical consistency and structural fidelity in reconstructions. I^2SB achieves superior quantitative performance, with root-mean-square error (RMSE) values of 49.8 HU on simulated noisy data and 152.0 HU on real data, outperforming state-of-the-art diffusion models such as conditional denoising diffusion probabilistic models (cDDPM) and patch-based diffusion methods. Moreover, its one-step inference enables reconstruction in just 0.19 s per 2D slice, representing over a 700-fold speedup compared to cDDPM (135 s) and surpassing DiffusionGAN (0.58 s), the second fastest. This combination of accuracy and efficiency indicates that I^2SB has potential for real-time or clinical deployment.

Quantitative Susceptibility Mapping (QSM) reconstructs tissue magnetic susceptibility from MR phase data but remains highly ill-posed in the single-orientation setting due to the cone-null region of the dipole kernel in the Fourier domain. To address this challenge, we propose QSMnet-INR, a physics informed framework that integrates an implicit neural representation (INR) into k-space modeling. The INR learns a continuous dipole response to improve stability in ill-conditioned regions, while a frequency-aware dipole loss enforces consistency with the physical model. Experiments on the 2016 QSM Reconstruction Challenge, a multi-orientation GRE dataset, and clinical data demonstrate improved reconstruction quality and reduced artifacts compared with existing methods, particularly under single orientation settings. Ablation and sensitivity analyses further support the complementary roles of INR-based modeling and frequency-aware regularization. While performance under more extreme susceptibility conditions or unseen acquisition settings warrants further investigation, the results indicate that integrating implicit representations with physics-informed constraints provides an effective approach for stabilizing ill-posed QSM reconstruction.

Brain stimulation is increasingly recognized as an effective and important therapeutic intervention for many brain diseases. Distance between the scalp and other brain regions is a pivotal variable for neurostimulation planning and the development of new techniques, but alterations in the distance between the scalp and other regions in brain diseases are largely unknown. In this study, we developed an automatic pipeline to calculate scalp-to-region distance (SRD) values from T1 MR images and applied it to a total of 1382 participants, including patients with autism spectrum disorder (ASD), Parkinson's disease (PD), Alzheimer's disease (AD), and cognitively normal controls (CNs). Cloud points were uniformly sampled on the automatically extracted scalp surface and cortex surface, on which the point-wise distance maps were generated. The brain was then coregistered with the BCI-DNI atlas, and SRD value for each brain region was extracted. Analysis of covariance (ANCOVA) was performed for SRD in each brain region, with age and sex as covariates. Compared with CNs, ASD patients showed widespread SRD decreases across the brain with prominent involvement of the frontal lobe, especially the orbitofrontal cortex and adjacent regions. In contrast, in AD patients, significantly increased SRD values were observed in various regions of the frontal gyrus. No significant SRD alteration was found in PD patients after correction. The automatic SRD calculation pipeline and the different patterns of SRD alterations in these diseases might be helpful for future neurostimulation planning in clinical practice.

Elucidating the mechanisms underlying brain aging and developing strategies to preserve or enhance cognitive abilities in aged rodents are central aims of both fundamental and translational neuroscience research. In this paper, we investigate the effects of red light stimulation (RLS) on enhancing learning and memory capabilities in aged rats, focusing on the interaction between the hippocampus, the anterior thalamus, and the dynamics of interstitial fluid (ISF) within the brain’s extracellular space (ECS). The experimental design included three groups: a light-treated group, a sham stimulation group, and a blank control group. Rats in the light-treated group were exposed to 638 nm red light for seven consecutive days. Additionally, electrodes were implanted in all groups to collect local field potential (LFP) data, and a water maze test was conducted to assess learning and memory capabilities. Behavioral results demonstrated that the light-treated group exhibited significantly improved efficiency in locating the hidden platform and recalling its spatial location in the water maze test. These cognitive enhancements were paralleled by increased ISF diffusion in the hippocampus, suggesting that RLS may facilitate better clearance of metabolic waste and improve neural communication within this region. LFP analysis revealed significant phase coupling synchrony between the hippocampus and thalamus following illumination, indicating a strengthened interaction between these regions. However, these effects diminished five days after the cessation of RLS exposure. In contrast, frequency coupling between the hippocampus and prefrontal cortex, as well as between the thalamus and prefrontal cortex, did not exhibit notable changes. These results suggest that the primary influence of RLS may be localized to the hippocampal-thalamic pathway, potentially mediated through enhanced ISF dynamics within the ECS. This research provides a substantial experimental foundation for the potential application of RLS to augment cognitive functions in the elderly and offers new insights into the role of the anterior thalamic nuclei and ECS in episodic memory, presenting valuable directions for future studies.

Magnetic stimulation has made significant strides in the treatment of psychiatric disorders. Nonetheless, current magnetic stimulation techniques lack the precision to accurately modulate specific nuclei and cannot realize deep brain magnetic stimulation. To address this, we utilized superparamagnetic iron oxide nanoparticles as mediators to achieve precise targeting and penetration. We investigated the effects of magnetic fields with varying frequencies on neuronal activity and compared the activation effects on neurons using a 10-Hz precise magneto-stimulation system (pMSS) with repetitive transcranial magnetic stimulation in mice. Oxytocin levels, dendritic morphology and density, and mouse behavior were measured before and after pMSS intervention. Our findings suggest that pMSS can activate oxytocinergic neurons, leading to upregulation of oxytocin secretion and neurite outgrowth. As a result, sociability was rapidly improved after a one-week pMSS treatment regimen. These results demonstrate a promising magneto-stimulation method for regulating neuronal activity in deep brain nuclei and provide a promising therapeutic approach for autism spectrum disorder.

Neurochemical imbalance is a contributing factor to neurological symptoms in multiple sclerosis (MS). The matured myelin sheath is crucial for substance transportation within the extracellular space (ECS) and for maintaining local homeostasis. Therefore, we hypothesize that disturbed ECS transportation following demyelinating lesions might lead to neurochemical imbalance in MS. In the current study, a lysophosphatidylcholine-induced unilateral MS model was used to investigate spatial neurochemical alterations. The results demonstrated that 168 substances were altered around the demyelination site in the ipsilateral hemisphere, compared to the contralateral hemisphere, with significant enrichment in the purine and arginine-proline metabolic pathways. Notably, dopamine was unexpectedly detected in the demyelinated region and the adjacent thalamus. Tracer-based MRI further revealed that the tracer injected into the striatum abnormally refluxed to the thalamus, with the area of reflux consistent with the altered dopamine distribution. The interstitial fluid drained extensively but was confined to the unilateral hemisphere, which may explain the observed widespread changes in other neuroactive substances. Importantly, after the restoration of ECS integrity, both interstitial fluid drainage and neurochemical imbalance, including dopamine, were normalized, supporting the potential link between ECS dysfunction and neurochemical imbalance. These observations highlight the crucial role of ECS transport in maintaining neurochemical homeostasis in the brain, providing new insights into the mechanisms that may underline the neuropsychiatric symptoms of MS.

Alzheimer's disease (AD) is characterized by cognitive and functional deterioration, with pathological features such as amyloid-beta (Aβ) aggregates in the extracellular spaces of parenchymal neurons and intracellular neurofibrillary tangles formed by the hyperphosphorylation of tau protein. Despite a thorough investigation, current treatments targeting the reduction of Aβ production, promotion of its clearance, and inhibition of tau protein phosphorylation and aggregation have not met clinical expectations, posing a substantial obstacle in the development of drugs for AD. Recently, artificial intelligence (AI), computational biology (CB), and systems biology (SB) have emerged as promising methodologies in AD research. Their capacity to analyze extensive and varied datasets facilitates the identification of intricate patterns, thereby enriching our comprehension of AD pathology. This paper provides a comprehensive examination of the utilization of AI, CB, and SB in the diagnosis of AD, including the use of imaging omics for early detection, drug discovery methods such as lecanemab, and complementary therapies like phototherapy. This review offers novel perspectives and potential avenues for further research in the realm of translational AD studies.

The transport of molecules within the brain extracellular space (ECS) plays a pivotal role in governing sleep patterns, memory formation, and the ageing process in organisms. Isolating and delineating the ECS is crucial for constructing accurate models of molecular dynamics within the intricate neuronal networks. However, the segmentation of the ECS has proven to be a formidable challenge due to its complex morphology, and there is a notable lack of comprehensive studies on such a topic. In this study, we have constructed an exclusive dataset for ECS segmentation, utilizing an advanced imaging technique enabled by cryo-electron microscopy. We introduce ECS-Net, a dedicated segmentation pipeline that integrates contrastive learning approach and shape-aware function. The contrastive learning strategy is employed to effectively differentiate the extracellular space from surrounding neural elements, while the shape-aware function is designed to bolster the model’s sensitivity to the distinctive structures of the ECS. The experimental results indicate that ECS-Net significantly outperforms existing methods in ECS segmentation, achieving an F1-score (F1) that is 2.91% higher and an Intersection-over-union (IOU) that is 4.62% superior on the ECSseg-1 dataset. Moreover, our model exhibits robust generalization capabilities when applied to an independent validation dataset. In summary, this study marks the establishment of the first cascaded network tailored for brain ECS segmentation. Our method is poised to enhance the analysis of ECS structures within an extensive dataset, thereby offering a substantial contribution to the broader biomedical research community.

Glioma is a highly lethal form of cancer, and its treatment requires overcoming several challenges, including the blood-brain barrier (BBB), rapid drug release at the tumor site, and monitoring of drug distribution. Herein, we have developed a bio-responsive optical/magnetic signal amplification nanoplatform (Gd/Cy5-Oxa@NPs) by conjugating a pH-responsive polymer block copolymer with cyanine 5 dye and then loaded this with gadolinium chloride and oxaliplatin (Oxa). The nanoplatform is administered via convection-enhanced delivery (CED) to hinder the progression of glioblastoma tumors. The CED of the nanoplatform offers a direct way to bypass the BBB, significantly reducing the required dosage without inducing detectable toxicity. The nanoplatform facilitates cell internalization and rapidly releases Oxa in acidic environments, enhancing treatment accuracy. In vivo experimental results demonstrate that the nanoplatform achieves remarkable anti-glioma efficacy through CED at low concentrations. Our research suggests that the combination of a pH-responsive optical/magnetic imaging nanoplatform and CED may represent a promising translational strategy for drug tracking and glioma treatment.