High-dimensional neuroimaging analyses for clinical diagnosis are often constrained by compromises in spatiotemporal fidelity and the limited adaptability of large-scale, general-purpose models. To address these challenges, we introduce Dynamic Curriculum Learning for Spatiotemporal Encoding (DCL-SE), an end-to-end framework centered on data-driven spatiotemporal encoding (DaSE). We leverage Approximate Rank Pooling (ARP) to efficiently encode three-dimensional volumetric brain data into information-rich, two-dimensional dynamic representations, and then decode this representation through a DGM-implemented architectural curriculum, where the fixed hierarchical decoder progressively refines features from global anatomical structures to fine pathological details without introducing a threshold-controlled stage-switch scheduler. Evaluated across six publicly available datasets, including Alzheimer’s disease and brain tumor classification, cerebral artery segmentation, and brain age prediction, DCL-SE consistently outperforms existing methods in accuracy, robustness, and interpretability. These findings underscore the critical importance of compact, task-specific architectures in the era of large-scale pretrained networks.
Medical Vision-Language Models (VLMs) are prone to hallucinations, compromising clinical reliability. While reinforcement learning methods like Group Relative Policy Optimization (GRPO) offer a low-cost alignment solution, their reliance on sparse, outcome-based rewards inadvertently encourages models to generate convoluted and unverifiable Chain-of-Thought reasoning to justify answers. This focus on outcomes obscures factual errors raising concerns about the reliability of such reasoning in medical applications. To address this, we propose CheXPO-v2, a novel alignment framework that shifts from outcome to process supervision. Our core innovation is a Knowledge Graph Consistency Reward mechanism driven by Entity-Relation Matching. By explicitly parsing reasoning steps into structured “Disease, Relation, Anatomy” triplets, we provide fine-grained supervision that penalizes incoherent logic and hallucinations at the atomic level. Integrating this with a hard-example mining strategy, our approach significantly outperforms answer-only GRPO and representative models on benchmarks like MIMIC-CXR-VQA. Crucially, CheXPO-v2 achieves an observed accuracy of 86.0% using only 5k samples, demonstrating exceptional data efficiency while improving reasoning fidelity and factual consistency. The project source code is publicly available at: https://github.com/ecoxial2007/CheX-Phi4MM.
The concurrency of ocular surface and fundus disorders is prevalent, posing a major threat to vision and quality of life. However, current therapies are difficult to simultaneously manage these coexisting symptoms. Here, we report a dual-location intervention strategy by in situ forming a permeable drug-carried bioactive nanocoating on ocular surface following instillation. The nanocoating is readily generated via dropping a mixture of polyphenol, metal ions, liposomes, and permeable drug. The coordination of metal ions with polyphenol and lipid enables nanocoating formation and its durable adhesion to ocular surface. The introduced bioactive polyphenol and liposomes separately provide anti-inflammatory activity and tear film-forming lipid to synergistically alleviate ocular surface disorder. In addition, the resulted network structure of the nanocoating offers continuous release of permeable drug, achieving intraocular delivery to mitigate the fundus symptom. Supported by the potency in synchronously intervening diabetic keratopathy and retinopathy in both streptozotocin-induced and Ins2-C96Y transgenic mouse models, this work opens a window for treating ocular diseases.
Repairing peripheral nerve injuries remains challenging because it requires biomaterials that can simultaneously modulate the adverse immune microenvironment and provide a bionic niche with integrated physical and biochemical cues. To address this, we developed a porcine sciatic nerve-derived acellular matrix membrane (SNCM) that uniquely preserved a native-like ensemble of extracellular matrix components (e.g., laminin, fibronectin, sulfated glycosaminoglycans) and endogenous neurotrophic factors. The SNCM exhibited superior physicochemical properties, including controlled degradation, enhanced mechanical robustness, high hydration, and favorable hemocompatibility. It also demonstrated potent immunomodulatory activity by efficiently scavenging reactive oxygen and nitrogen species. In vitro, SNCM specifically promoted neuronal differentiation of PC12 cells and enhanced Schwann cell migration, proliferation, and pro-regenerative phenotype activation. In a rabbit tibial nerve injury model, SNCM application significantly reduced the incidence of severe perianastomotic adhesion and neuroma formation (with no severe adhesion cases detected in the SNCM cohort), substantially improved functional recovery (evidenced by increased muscle wet weight ratio), and facilitated structural regeneration characterized by orderly axonal regrowth, coordinated Schwann cell activity, and robust remyelination, achieving a near-complete myelinated nerve fiber regeneration rate. By synergistically integrating structural, neurotrophic, and immunomodulatory functions into a single biomimetic platform, SNCM actively orchestrates a pro-regenerative microenvironment, positioning it as a highly competitive and clinically translatable strategy for advanced nerve repair.
Activating PIK3CA mutations contribute to lesion growth in a subset of cerebral cavernous malformation (CCM) cases. STX-478 is an allosteric, mutant-selective PI3Kα inhibitor, but its activity in CCM is unclear. We evaluated STX-478 in a mouse model driven by combined endothelial expression of PIK3CA(H1047R) and MAP3K3(I441M). At postnatal day 5 (P5), C57BL/6J pups were injected with endothelial-tropic AAV-BR1 vectors encoding these mutations or with control AAV. Baseline MRI was obtained at P14 before treatment: mice received STX-478 (20 mg/kg intravenously every other day) or vehicle for two weeks, yielding four experimental groups. MRI and tissue analyses were performed at P28. Compared with vehicle-treated model mice, STX-478-treated model mice had lower summed susceptibility-weighted imaging (SWI) focus counts, smaller histological lesion areas, reduced hemosiderin deposition, and lower phosphorylated AKT (p-AKT) fluorescence intensity. Body weight increased in all groups during treatment. These findings support attenuation of the experimental CCM phenotype after baseline assessment, without establishing regression of individual lesions. Reduced hemosiderin deposition represents a haemorrhage-related histological finding rather than a direct measure of haemorrhage frequency. This study provides a preclinical rationale for further evaluation of mutant-selective PI3Kα inhibition in PIK3CA-mutant CCM, including dose optimisation, pharmacokinetic assessment, longer-term safety studies, and validation in additional models.
