The human brain is an intricate network structurally organized across multiple scales. Multiscale structural architecture arises from complex biological mechanisms and provides the anatomical substrate for functional interactions. However, the extent to which the multiscale structural connectome constrains functional activity remains unclear. Here, we investigate the structure-function relationship by constructing an in vivo multiscale structural connectome that integrates white matter tractography, microstructural similarity, and cortico-cortical proximity. Multiscale structural connectome eigenmodes outperform conventional approximations of structural connectivity in capturing spontaneous and task-evoked functional activity. Moreover, multiscale structure-function decoupling reveals an organizational axis spanning from coupled unimodal sensory to decoupled transmodal association cortices, recapitulating microstructure and macroscale functional hierarchies. These decoupled patterns spatially align with opioid neurotransmitter receptors and mitochondrial succinate dehydrogenase, and colocalize with transcriptomic signatures enriched in synaptic structure and signaling regulation. Collectively, our findings highlight the pivotal role of multiscale structural wiring in shaping brain functional dynamics and provide novel neurobiological insights into the structure-function relationship.

BACKGROUND:The glymphatic system, a key fluid clearance pathway in the central nervous system, is emerging as a potential therapeutic target for synucleinopathies. Dysregulation of this system may contribute to spinocerebellar ataxia type 3 (SCA3) pathogenesis, in which the accumulation of misfolded proteins acts as a central driver. OBJECTIVES:The goal was to investigate glymphatic system function in SCA3 patients and evaluate its relationship with brain damage and clinical disability. METHODS:Ninety-two SCA3 patients (14 with premanifest SCA3 and 78 with manifest SCA3) and 98 healthy controls underwent clinical evaluation and magnetic resonance imaging (MRI) scans. MRI parameters, including the diffusion along the perivascular space (DTI-ALPS) index (a proxy for glymphatic function); cerebral, cerebellar, and subcortical gray matter volumes; and whole-brain microstructural properties of white matter, were calculated. RESULTS:Patients with premanifest and manifest SCA3 had lower ALPS indices compared with healthy controls, and patients with manifest SCA3 had a lower ALPS index compared with premanifest SCA3. In SCA3 patients, lower ALPS index was associated with more severe disability and longer disease duration. A negative correlation between ALPS and disease duration emerged after 3 years, with no significant association observed before the 3-year cutoff. Moreover, lower ALPS index was correlated with more pronounced cortical and subcortical gray matter atrophy, decreased fractional anisotropy, and elevated mean diffusivity in white matter. CONCLUSIONS:Our findings demonstrate that glymphatic function is impaired, particularly in the presymptomatic stage of SCA3, and this impairment is associated with disability, neurodegeneration, and demyelination. Therefore, glymphatic dysfunction may contribute to the pathogenesis of SCA3. © 2026 International Parkinson and Movement Disorder Society.

Schizophrenia has been linked to reduced cortical thickness and abnormal gene expression. While antipsychotic treatment has been found to affect cortical morphology and gene expression, its impact on subject-specific deviations in cortical morphometric similarity and the underlying genetic mechanisms remain unclear. To quantify risperidone-related changes in morphometric similarity at the individual level and test their spatial alignment with cortical transcriptomic patterns. Twenty-four drug-naive first-episode schizophrenia patients and 30 healthy controls underwent T1-weighted imaging scans. Patients were scanned before and after 12 weeks of treatment with risperidone; symptoms and cognitive function were assessed with PANSS and MCCB scale. For each scan, cortical morphometric similarity matrices were built from regional cortical thickness distributions using the Wasserstein distance. We defined Morphometric Similarity Deviation (MSD) as a subject-level, normative-referenced departure from the healthy morphometric similarity pattern, derived from node-wise fingerprint correlations with the healthy template. Partial least squares regression related treatment-induced MSD changes to cortical transcriptomic data obtained from the Allen Human Brain Atlas. Patients exhibited high MSD in the frontal, temporal, and temporoparietal regions. Greater baseline MSD across the whole brain and multiple networks were associated with more severe positive symptoms. After treatment, MSD decreased, and reductions within the salience/ventral attention network associated with improved Emotional Intelligence. Moreover, risperidone-induced changes in MSD were spatially correlated with the expression of specific genes enriched in neurotransmission, cell adhesion, immune function, and schizophrenia. Specific expression analyses revealed that these genes were specifically expressed in astrocytes and oligodendrocytes, and spanned almost all developmental stages. Risperidone reduces MSD, reflecting convergence toward a normative cortical morphometric similarity pattern. These changes were spatially aligned with gene expression patterns involved in neurotransmission and immune processes, suggesting a molecular basis for treatment-linked structural normalization and its cognitive benefits.

Brain white matter (WM) has traditionally been viewed as a passive conduit for neural transmission. However, evidence of blood oxygen level-dependent (BOLD) signals measured from the WM suggests its active participation in grey matter (GM) functional networks. Using 7-Tesla functional MRI (fMRI) data, we constructed a GM-WM functional connectome. We found that GM-WM functional architecture follows the unimodal-transmodal hierarchy of GM and is shaped by distributions of neurotransmitter receptors. Distinct WM networks exhibit unique connectivity profiles with GM, reflecting their roles in specific cognitive domains. Individual variations in this connectome correlated with cognitive performance. Notably, compared with the traditional GM-GM functional connectome, the GM-WM functional connectome shows stronger associations with brain disorders, suggesting greater diagnostic sensitivity as a neuromarker. These findings are replicated in a 3-Tesla fMRI cohort. Our work establishes WM as an integral component of the brain's functional architecture, contributing to hierarchical architecture and supporting higher-order cognition.

Robust assistive BCIs require the hierarchical decoding of coarse-grained context and fine-grained object semantics from neural activity, yet these signals can interfere when decoded jointly. However, whether these semantic levels interact competitively or synergistically within neural representations remains a fundamental question. Using high-resolution 7T fMRI from the Natural Scenes Dataset (NSD), we built a unified multi-task decoder that reconstructs scene categories, multi-label object semantics, and natural-language captions from visual cortical activity under a controlled single/dual/multi-task comparison protocol. This unified approach allowed for a controlled quantification of cross-task interactions. We identify a “global-local interference” effect: joint decoding of categories and object labels produces consistent mutual performance degradation, suggesting representational competition that can undermine multi-command control. Clinically, this interference represents a safety-relevant bottleneck for assistive BCIs, where concurrent high-level navigation intent and object-level manipulation intent may degrade together and destabilize device control. Crucially, adding linguistic supervision functions as a top-down semantic regularizer that mitigates interference and improves robustness of the decoded control-relevant representations. This framework offers a pivotal solution for next-generation neuroprosthetics, resolving the trade-off between scene awareness and object interaction. By bridging the gap from isolated command recognition to holistic interpretation, it enables safer, context-aware control of assistive devices and restores coherent narrative communication for patients with aphasia or locked-in syndrome.

Resting-state functional and structural brain imaging are considered the foundation for understanding task-related brain activation patterns. However, few studies have integrated multidimensional functional and structural brain features into predictive models for predicting task-specific functional activation. To address this, we propose PG-DCAM, which simultaneously integrates resting-state functional and structural features to achieve precise predictions of individual task-related brain activation patterns. First, the model employs a dual-channel attention network to merge functional and structural brain features, extracting both global and local information. Second, a prompt-guided mechanism is introduced to enhance the deep interaction between brain information and task information, allowing the model to focus on key features when predicting different tasks. Finally, we adopt a Categorized-Contrastive learning strategy to address the challenge of task differentiation in multi-task learning. This strategy not only improves prediction accuracy but also substantially enhances training efficiency. Experiments confirm that our model accurately predicts individual task-related brain activation patterns from the HCP dataset, achieving state-of-the-art (SOTA) performance in the field. This study presents a comprehensive feature integration model that offers a novel approach to predicting individual cognitive traits and contributing to a deeper understanding of human cognition mechanisms.

Personalized visual neural encoding is limited by inter-individual variability in brain structure and function, which conventional stimulus-driven models often ignore. We propose Indiv-VEM, a subject-aware framework that integrates sMRI, rsfMRI, natural images, and text semantics. It derives individualized brain representations, refines them by frequency-domain fusion, and aligns them with global semantics and local stimulus details through spatial-frequency cross-attention. On NSD, Indiv-VEM outperforms linear, graph-based, and attention baselines in voxel-wise fMRI prediction and better preserves inter-subject response patterns. Ablations confirm the contributions of individualized neuroimaging features, frequency-domain fusion, and dual-domain attention. The framework offers a foundation for personalized functional mapping, presurgical planning, and precision neurostimulation.

Decoding the brain’s visual neural activity is crucial for understanding visual mechanisms and advancing brain-computer interface (BCI) technology. Existing methods often rely on static alignment or non-aligned strategies, making it difficult to fully utilize the semantic information in brain activity, resulting in suboptimal decoding performance, especially in complex tasks. To address this, we propose a Dynamic Aligned Visual Decoding Model (DA-VDM), based on a generative language model, employing a dynamic alignment strategy implemented via progressively weighted training. By dynamically adjusting the training weights, this strategy gradually shifts the model’s focus from low-level feature mapping to high-level semantic generation, thereby enhancing the semantic association between brain activity features and image-text representations and improving semantic expression capabilities. Additionally, we introduce Prompt-based techniques, leveraging the generative power of language models to tackle cross-subject and cross-task decoding challenges, enabling the model to adapt to different subjects within a unified framework while predicting both the category and textual description of visual stimuli. Experiments on the Natural Scenes Dataset (NSD) demonstrate that DA-VDM achieves an accuracy of 0.685 in category decoding tasks, outperforming existing models; in text decoding tasks, it also leads in metrics such as METEOR and ROUGE. Ablation studies confirm that the dynamic alignment strategy significantly enhances category and text decoding accuracy, while Prompt-based techniques exhibit strong generalization capabilities in cross-subject and cross-task decoding.

BACKGROUND:Cognitive and behavioral symptoms of major depressive disorder (MDD) are linked to aberrant changes in the controllability of brain networks. However, previous studies examined network controllability using white matter tractography, neglecting the contributions of gray matter. We aimed to examine differences in the controllability of morphometric networks between patients with MDD and demographic-matched healthy controls and identify the associated neurobiological signatures. METHODS:Based on the structural and diffusion MRI data from two independent cohorts, we calculated the controllability of morphometric similarity networks for each participant. A generalized additive model was used to investigate the case-control differences in regional controllability and their cognitive and behavioral associations. We investigated the associations between imaging-derived controllability and neurotransmitters, brain metabolism, and gene transcription profiles using multivariate linear regression and partial least squares regression analyses. RESULTS:In both cohorts, depression-related abnormalities of morphometric network controllability were primarily located in the prefrontal, cingulate, and visual cortices, contributing to memory, sensation, and perception processes. These abnormalities in network controllability were spatially aligned with the distributions of serotonergic transmission pathways as well as with altered oxygen and glucose metabolism. In addition, these abnormalities spatially overlapped with differentially expressed genes enriched in annotations related to protein catabolism and mitochondria in neuronal cells and were disproportionately located on chromosome 22. CONCLUSIONS:Collectively, neuroimaging evidence revealed aberrant morphometric network controllability underlying MDD-related cognitive and behavioral deficits, and the associated genetic and molecular signatures may help identify the neurobiological mechanisms underlying MDD and provide feasible therapeutic targets.

The human brain is an extraordinarily complex spatio-temporal network system. The attention mechanism in graph neural networks (GNNs) has demonstrated promising performance in learning brain network representations. However, existing GNN methods with such attention mechanisms only aggregate information from neighboring brain nodes in each layer and ignore indirect node connections, limiting the graph modeling capability. Here, we propose a multi-hop spatio-temporal graph convolutional network (MSTGCN) with reverse contrastive (RevCon) learning. Specifically, MSTGCN employs a novel diffusion process that considers all paths between unconnected brain nodes, thereby incorporating multi-hop contextual information into each layer of GNN attention computation. Additionally, a temporal attention module is designed to extract dynamic functional connections and aggregate features into a dynamic graph-level representation. Finally, we adopt the RevCon learning strategy as a regularization mechanism to improve model generalizability under heterogeneous multi-site neuroimaging data by discouraging over-reliance on site-specific representations and promoting disease-relevant features. We validate the performance of the proposed model on diagnostic and prognostic tasks in two representative brain network disorders. Experiments on two private independent epilepsy datasets and the public ABIDE dataset demonstrate the effectiveness of MSTGCN, achieving identification accuracies of up to 85.52%, 78.27%, and 69.23%, respectively, and a cross-site classification accuracy of 82.07%, outperforming several state-of-the-art methods. Furthermore, our model achieves an accuracy of 82% in predicting surgical outcomes. The interpretability results of the model are consistent with previous medical studies. In summary, MSTGCN is a novel model for characterizing brain networks and provides insights into diagnosis and prognosis of brain disorders.

Integrating multimodal semantic features, such as images and text, to enhance visual neural representations has proven to be an effective strategy in brain visual decoding. However, previous studies have either focused solely on unimodal enhancement techniques or have inadequately addressed the alignment ambiguity between different modalities, leading to an underutilization of the complementary benefits of multimodal features or a reduction in the semantic richness of the resulting neural representations. To address these limitations, we propose a Multimodal Fusion Alignment Neural Representation Model (MFA-NRM), which enhances visual neural decoding by integrating multimodal semantic features from images and text. The MFA-NRM incorporates a fusion module that utilizes a Variational Autoencoder (VAE) and a self-attention mechanism to integrate multimodal features into a unified latent space, thereby facilitating robust semantic alignment with neural activity. Additionally, we introduce prompt techniques that adapt neural representations to individual differences, improving cross-subject generalization. Our approach also leverages the semantic knowledge from ten large pre-trained models to further enhance performance. Experimental results on the Natural Scenes Dataset (NSD) show that, compared to unimodal alignment methods, our method improves recognition tasks by 18.8% and classification tasks by 4.30%, compared to other multimodal alignment methods without the fusion module, our approach improves recognition tasks by 33.59% and classification tasks by 4.26%. These findings indicate that the MFA-NRM effectively resolves the problem of alignment ambiguity and enables richer semantic extraction from brain responses to multimodal visual stimuli, offering new perspectives for visual neural decoding.

Humans develop shared concepts of others' emotions to support adaptive social functioning, yet how these concepts are dynamically represented in major depressive disorder (MDD) during naturalistic movie viewing is not yet fully established. Using functional MRI, we examined patients with MDD (n = 55) and healthy controls (HCs; n = 62) as they freely viewed movie clips depicting happy and sad emotions. Neural similarity was quantified with inter-subject correlation at whole-brain, network, and regional levels, and its association with emotional traits was assessed using inter-subject representational similarity analysis. Compared with HCs, patients with MDD showed significantly reduced whole-brain similarity, particularly during sad contexts. Network analyses revealed that HCs exhibited increased similarity in the limbic network during sadness, reflecting a shared "sadness resonance," whereas patients with higher depressive severity showed widespread disruptions across visual, limbic, dorsal attention, and default mode networks. At the regional level, similarity in the inferior temporal gyrus and lateral occipital cortex was closely linked to individual differences in emotional awareness, with pronounced context- and region-specificity. These findings highlight neural decoupling and heterogeneity as core features of MDD and provide new evidence for potential biomarkers to inform risk assessment and personalized interventions.

Background: Schizophrenia is a highly heritable mental disorder associated with widespread anatomical alterations during neurodevelopment. Converging evidence suggests transcriptomic architecture underlying brain abnormalities in schizophrenia, while how individualized brain morphological deviations relate to gene expression levels remains unclear. Methods: To investigate individual-level brain deviations and its transcriptomic signatures in schizophrenia, this study collected T1-weighted MRI data from 95 early-onset schizophrenia (EOS) patients and 99 typically developing (TD) controls. Normative modeling was used to measure individual deviations in cortical thickness and subcortical volume. Partial least squares regression was calculated to capture covarying patterns between structural deviations and whole-brain gene expression levels. Clustering analysis was performed on latent brain-gene covarying components, and the results were further functionally decoded through gene enrichment analyses. Results: Group-level comparisons suggested patients with EOS showed consistently decreased z-scores of cortical thickness in the frontal and temporal lobe regions, while increased inter-individual variability in the lingual gyrus. Clustering analysis of z-scores with transcriptomic signatures identified two distinct brain-gene covarying subtypes. Subtype 1 showed thickening cingulate gyrus, thinning occipital pole, and atrophic subcortical nuclei. Subtype 2 exhibited widespread cortical thinning across the frontal, parietal, temporal, and limbic regions, but enlarged subcortical nuclei. Genes underlying two subtypes were both enriched for neurodevelopmental diseases. However, subtype 1 was associated with synaptic transmission, and subtype 2 was related to cytoskeletal and neuronal connectivity. Conclusion: This study reveals individual-level anatomical deviations and transcriptomic heterogeneity in early-onset schizophrenia. The findings provide an individualized brain-gene coupling framework for understanding pathophysiology of schizophrenia during brain development. ### Competing Interest Statement The authors have declared no competing interest.

Background and objective: Despite the potential of artificial intelligence in neuro-oncology, its clinical translation remains constrained by high computational overhead and the challenges associated with 3D medical image analysis. This study aims to develop an accessible diagnostic platform that facilitates efficient brain tumour detection and localization on conventional computing systems without the need for specialized technical expertise or complex installation procedures. Methods: We present MediScreen-Brain, a lightweight, clinically oriented software platform. To overcome the computational challenge of transitioning from two-dimensional to threedimensional image analysis, we propose an anatomy-guided rapid localization (AGRL) algorithm. This algorithm uses the natural structural continuity of the brain to narrow the search area, thereby avoiding unnecessary processing of every slice in the volume. The system was trained on public datasets and validated using a private clinical cohort of 561 MRI scans covering meningiomas, gliomas, and pituitary tumours. Results: Compared with conventional sequential slice-wise processing, the AGRL algorithm reduced 3D MRI processing time by 76.8% while maintaining 89.1% localization consistency. The platform also automatically quantifies tumour dimensions and volume, generating structured reports that integrate seamlessly into existing clinical workflows. Conclusion: MediScreen-Brain effectively connects advanced image analysis with practical clinical needs. Its hardware-agnostic design and zero-configuration architecture provide a viable solution for both well-equipped and resource-limited medical facilities. This platform enhances the feasibility of rapid 3D brain tumour assessment and significantly advances the translation of AI-driven diagnostic tools in neuro-oncology. The source code and software are publicly available at https://jingw-ui.github.io/MediScreen-Brain/.

Neuroimaging research depends on heterogeneous software, multimodal data and multistage statistical workflows. Large language model (LLM)-based agents offer a route to automate these workflows, but their susceptibility to hallucination limits their credibility in scientific use. Here we introduce NEURA, a proof-carrying framework for hallucination-resistant neuroimaging automation. NEURA converts free-text research questions and neuroimaging datasets into executable analysis plans, validated outputs and structured reports. The system combines disease- and tool-aware planning with a deterministic verification layer inspired by formal proof: before any claim is retained for reporting, it must be checked against tool-derived evidence and domain axioms. On NeuroEval, an expert-curated benchmark of 110 neuroimaging tasks, NEURA achieved 89.5% planning accuracy, a 30.5% improvement over direct LLM queries. In a controlled hallucination-injection experiment, the verification layer detected all the injected error classes under the specified axiom bank and trust assumptions, with no false positives. In case studies of spinocerebellar ataxia type 3, NEURA reproduced cerebellar atrophy and abnormal diffusion patterns consistent with established pathology and independent expert analyses. Together, these findings show that coupling domain-grounded agency with proof-carrying verification can turn LLM-driven workflow automation from probabilistic self-checking into auditable scientific computation.

Brain decoding, which aims to infer cognitive states from recorded neural activity, plays a critical role in understanding cortical information processing. Recent decoding studies have made significant advances in single-trial decoding by leveraging multimodal MRI data, enabling models to move beyond capturing local brain responses to integrating global brain information. However, these studies typically assume a fixed fMRI input length and a static brain connectome for each trial, which conflicts with the inherently time-varying nature of the brain. Here, we propose an Adaptive Structure-Function Fusion Network (ASFFNet) that decodes cognitive states in a dynamic manner. Firstly, ASFFNet learns trial-specific functional connectomes through hierarchical attention modules and integrates them with prior structural connectomes via a hemodynamics-informed adaptive mechanism. Secondly, the integrated connectomes are used to aggregate whole-brain features, followed by an LSTM-based classifier for decoding. Lastly, ASFFNet dynamically modulates its input window for each trial using a confidence-driven inference strategy. We evaluate ASFFNet using task-fMRI data from the Human Connectome Project database, involving a large cohort of 1200 participants across 21 distinct cognitive states. ASFFNet achieved an average decoding accuracy of 80.31% across different input windows and an information transfer rate of 47.29 bits/min under the dynamic inference setting, both significantly outperforming the benchmark methods. Notably, ASFFNet can extract neural representations resembling task-relevant brain activations, and its confidence levels distinguish easy trials from difficult ones. This study advances brain decoders from fixed-length to adaptive-length inputs and has substantial potential to improve the efficiency of brain-computer interfaces.

Background Early-onset schizophrenia (EOS) emerges during brain maturation, yet its neuropathologic mechanisms remain poorly understood. Based on the concept that pathological disturbances may propagate via the brain connectome, this study aims to delineate EOS cortical maturation abnormalities using a network diffusion model (NDM). Method Cortical thickness deviations were computed for 95 EOS patients (ages 7–17) and 99 controls. Constrained by a multimodally-fused microscale normative connectome, the study identified pathological epicenters. A sliding window approach and NDM were employed to characterize the dynamic transition and spread of these abnormalities across ages. Result Constrained by microscopic connectomes, cortical thinning epicenters consistently localized in the higher-order association cortices (frontal, parietal, and temporal lobes), while thickening epicenters in the sensorimotor and visual networks. The NDM effectively predicted abnormalities in adjacent age groups, indicating that pathology orderly spread along the connectome. Disease epicenters transferred in two pathways: thickening epicenters consistently located in sensory regions from childhood to adolescence; thinning epicenters migrated from the frontal lobe to the parietal lobe. Conclusion This study reveals microscopic connectomic-limited disease epicenters in EOS and characterizes their pathological propagation progression during brain maturation. Broadly, these findings provide a mechanistic framework that may aid in clinical intervention of schizophrenia.

Aberrant dynamic shifts in brain states are a hallmark of cognitive and behavioral dysfunctions in major depressive disorder (MDD), yet the underlying mechanisms of these disturbances remain elusive. Leveraging network control theory of morphological networks, we characterized aberrant brain dynamics and energy deficits of MDD patients in two independent cohorts. MDD patients exhibited reduced dynamic stability, characterized by elevated intra-state transitions and diminished inter-state transitions, which were associated with impaired control energy. Region-specific deficits of energy regulation capacity were observed in key nodes of the default mode and limbic networks, including the posterior cingulate cortex and temporal pole, which correlated with cognition and clinical symptoms in MDD patients. MDD-related energy inefficiency was related to multiscale energy architectures at cellular, molecular, and biological levels, including mitochondrial morphologies and functions, energy metabolism pathways, and brain metabolic patterns. Additionally, we demonstrated an association between energy demands and cortical dynamics, indicating a disrupted energy-dependent neurophysiological activity in MDD patients. Together, these results identified the energetic fundamentals underlying pathological brain-state transitions in MDD patients. Identifying energy-vulnerable nodes from a controllability perspective may therefore provide valuable targets for restoring normative neural dynamics in MDD.

Background: Peripheral inflammation is implicated in the pathophysiology of schizophrenia, but how inflammatory signals map onto the large-scale brain organization remains incompletely understood. Methods: We applied a supervised multimodal fusion approach guided by interleukin-6 (IL-6) to gray matter volume (GMV) and resting-state regional homogeneity (ReHo) from a population-based discovery cohort in the UK Biobank. Brain components related to IL-6 were identified and then projected onto an independent schizophrenia cohort to examine their relevance to the disease. Imaging-transcriptomic analyses using the Allen Human Brain Atlas characterize the molecular substrates underlying the disease-relevant pattern. Results: Two ReHo components were significantly associated with plasma IL-6, while no GMV components showed robust IL-6 correlations. One of the components (ReHo IC4) exhibited a conserved functional pattern characterized by enhanced visual synchrony and reduced synchrony in the medial prefrontal cortex. This pattern remained unchanged in both the healthy controls and patients. In contrast, another component (ReHo IC8) showed increased synchrony in the default mode network and reduced synchrony in sensorimotor networks, and its loadings were significantly elevated in patients with schizophrenia. Imaging-transcriptomic analysis revealed the molecular architecture of this disease-amplified pattern. The default mode region was enriched in synaptic signaling pathways, while the sensorimotor region was linked to mitochondrial bioenergetic processes; both patterns significantly enriched with gene sets related to schizophrenia. Conclusions: This study identified an IL-6-associated functional brain pattern that is amplified in schizophrenia, linking peripheral inflammation to disease-specific network dysregulation. The findings provide a systems-level framework for understanding how peripheral inflammation interacts with large-scale brain network activities in schizophrenia. ### Competing Interest Statement The authors have declared no competing interest.

Spinocerebellar ataxia type 3 (SCA3) is a rare neurodegenerative disorder characterized by ataxia; structural and functional damage to the cerebrocerebellar loop play key roles in its pathology. However, effective treatments for SCA3 remain limited. Repetitive transcranial magnetic stimulation (rTMS) modulates cortical plasticity. Here, we investigated the utility of rTMS in SCA3 treatment. This study included 25 confirmed SCA3 patients and 33 age- and sex-matched healthy volunteers as controls. The Scale for the Assessment and Rating of Ataxia (SARA) and the International Cooperative Ataxia Rating Scale (ICARS) were used to assess the severity of clinical symptoms in the SCA3 group. Both groups completed neuropsychological evaluations and underwent brain magnetic resonance imaging (MRI) before and after treatment. MRI data were preprocessed using DPABI software to analyze changes in functional connectivity strength, both at the stimulation target and across the whole brain, in SCA3 patients before and after multi-target rTMS therapy based on the cerebrocerebellar loop. This clinical study was registered on October 28, 2020, registration number ChiCTR2000039434. After multi-target rTMS treatment, SARA (p < 0.001) and ICARS (p < 0.001) scores in SCA3 patients were significantly reduced, whereas Montreal Cognitive Assessment (p < 0.001) scores showed a substantial improvement in cognitive performance. Functional connectivity strengths between the paracentral lobule and cerebellum, and between the cerebellar vermis and paracentral lobule, decreased in SCA3 patients after treatment, gradually approaching levels observed in healthy controls. A multi-target rTMS treatment strategy targeting the cerebrocerebellar loop may significantly improve motor and cognitive functions in SCA3 patients by effectively regulating functional connectivity within this circuit.