White matter hyperintensity (WMH) is a primary manifestation of small vessel disease (SVD), leading to vascular cognitive impairment and other disorders. Accurate WMH quantification is vital for diagnosis and prognosis, but current automatic segmentation methods often fall short, especially across different datasets. The aims of this study are to develop and validate a robust deep learning segmentation method for WMH of vascular-origin. In this study, we developed a transformer-based method for the automatic segmentation of vascular-origin WMH using both 3D T1 and 3D T2-FLAIR images. Our initial dataset comprised 126 participants with varying WMH burdens due to SVD, each with manually segmented WMH masks used for training and testing. External validation was performed on two independent datasets: the WMH Segmentation Challenge 2017 dataset (170 subjects) and an in-house vascular risk factor dataset (70 subjects), which included scans acquired on eight different MRI systems at field strengths of 1.5T, 3T, and 5T This approach enabled a comprehensive assessment of the method's generalizability across diverse imaging conditions. We further compared our method against LGA, LPA, BIANCA, UBO-detector and TrUE-Net in optimized settings. Our method consistently outperformed others, achieving a median Dice coefficient of 0.78 ± 0.09 in our primary dataset, 0.72 ± 0.15 in the external dataset 1, and 0.72 ± 0.14 in the external dataset 2. The relative volume errors were 0.15 ± 0.14, 0.50 ± 0.86, and 0.47 ± 1.02, respectively. The true positive rates were 0.81 ± 0.13, 0.92 ± 0.09, and 0.92 ± 0.12, while the false positive rates were 0.20 ± 0.09, 0.40 ± 0.18, and 0.40 ± 0.19. None of the external validation datasets were used for model training; instead, they comprise previously unseen MRI scans acquired from different scanners and protocols. This setup closely reflects real-world clinical scenarios and further demonstrates the robustness and generalizability of our model across diverse MRI systems and acquisition settings. As such, the proposed method provides a reliable solution for WMH segmentation in large-scale cohort studies.

Sphingolipids are essential, complex lipids that are abundant in the cell membranes of eukaryotic cells, particularly concentrated in the myelin and neuronal membranes of the central nervous system (CNS). These lipids are crucial components of the cell membrane, affecting their structure and fluidity, and thus regulating various biological processes, including signal transduction, cell differentiation, apoptosis, and autophagy. The metabolic pathways of sphingolipids are highly complex and conserved, and this metabolic process can produce multiple metabolites. Metabolites such as ceramide (Cer) and sphingosine‐1‐phosphate (S1P) are vital in CNS signaling, affecting neurodevelopment, myelination, and synaptic plasticity. Thus, disruption of sphingolipid metabolism is closely related to neurological disorders. This article provides the latest studies concerning the known features of sphingolipid and sphingolipid metabolism, highlighting its physiological and pathological roles in the CNS.

Medical image segmentation serves as a fundamental step in computer-aided diagnosis and treatment planning. Despite the remarkable progress achieved by fully supervised deep learning models, their performance heavily depends on large-scale, high-quality annotated datasets, which are labor-intensive and expensive to acquire. Semi-supervised learning (SSL) mitigates this issue by leveraging both limited labeled data and abundant unlabeled data. However, existing SSL approaches in medical imaging often lack effective semantic modeling and suffer from domain gaps between natural image pretraining and medical domains.In this work, we present DAVLMF-Seg, a novel Domain-Adaptive Vision-Language Model (VLM) guided Frequency-aware semi-supervised segmentation framework that fully exploits VLM's cross-modal semantic alignment. Through parameter-efficient domain-adaptive fine-tuning, DAVLMF-Seg aligns medical images and text descriptions within a shared latent space to provide prior-aware guidance during pseudo-label refinement. Additionally, we propose a frequency-domain conditioned diffusion architecture that progressively fuses semantic features and mitigates decoding ambiguity. To address confidence calibration bias, we further introduce an uncertainty-aware regularization strategy based on multi-step diffusion outputs. Extensive experiments on multiple benchmark datasets demonstrate that our approach achieves superior performance over state-of-the-art SSL methods, particularly under extremely limited supervision and in cross-domain generalization tasks.

Vision–language models (VLMs) offer a promising mechanism for injecting semantic priors into medical image segmentation. However, directly adapting VLMs pretrained on natural image–text pairs remains challenging due to insufficient cross-modal alignment, weak semantic grounding, and limited utilization of fine-grained clinical descriptions. This study aims to enhance semantic alignment and spatial discrimination in vision–language-based medical image segmentation. We propose a vision–language segmentation framework that exploits fine-grained textual supervision and constrained cross-modal interaction. A soft multi-positive vision–language contrastive learning objective is introduced to explicitly model the one-to-many correspondence between each medical image and multiple clinical textual descriptions, enabling both global and local cross-modal alignment. To prevent representation drift toward overly abstract textual semantics, a high-level semantic alignment objective is further designed, incorporating structural consistency constraints, relative similarity preservation, and orthogonality regularization. In addition, a multi-latent attention-based cross-modal fusion module is developed to integrate vision–language priors into the segmentation backbone via modality-aware gating and latent aggregation. Experiments on multiple public medical image segmentation benchmarks demonstrate that the proposed method consistently outperforms existing vision–language and vision-only approaches in segmentation accuracy and generalization performance. The proposed framework effectively improves semantic grounding and cross-modal alignment for medical image segmentation. This work provides a robust and principled strategy for adapting vision-language models to medical image segmentation, facilitating more reliable exploitation of clinical semantic knowledge.

Background: White matter hyperintensity (WMH) is a hallmark of cerebral small vessel disease and an important contributor to vascular cognitive impairment (VCI), yet lesion burden incompletely explains interindividual variability in cognitive outcomes across the VCI continuum. Functional connectome signatures relevant to this variability remain incompletely characterized. Methods: We analyzed multicenter resting-state functional MRI data from 247 participants spanning vascular risk factors with normal cognition, vascular mild cognitive impairment, and vascular dementia. Exploratory external testing was performed in an independent dataset of 37 participants. Connectome-based predictive modeling (CPM) with permutation testing was used. To reduce circularity, we computed cross-validated network strength (cvNS) using predictive masks defined within training folds only. Results: We identified cross-validated functional connectivity patterns associated with MoCA (positive model, p_perm = 0.034) and TMT-B performance (negative model, p_perm = 0.001). These patterns were most prominently represented in Frontoparietal, Motor, and Subcortical–Cerebellar regions. The TMT-B negative network showed a substantial contribution from weak between-network connections (64.4% of predictive edges). In the external dataset, network strength computed from discovery consensus masks remained associated with MoCA. Greater WMH volume was associated with worse TMT-B performance, and mediation analyses indicated that cvNS computed from the TMT-B connectivity pattern statistically linked WMH volume to both TMT-B (indirect = 0.090, 95% CI [0.018, 0.225], p = 0.023) and MoCA performance (indirect = −0.007, 95% CI [−0.019, −0.001], p = 0.031). Conclusions: CPM-derived functional connectivity patterns capture meaningful continuous variability in cognition across the VCI continuum and provide statistical support consistent with WMH-related functional disconnection as a network-level correlate of cognitive impairment.