福建医科大学

Accurate and generalizable pathological segmentation is essential for clinical deployment. However, existing methods remain challenged by cross-center domain shifts and ambiguous tissue boundaries, often relying on spurious domain-specific correlations and producing unreliable predictions in structurally complex regions. To address these issues, we propose the Causal-Evidential Generalization Network (CaEG-Net) for pathological image segmentation, which draws on ideas from causal inference and Dempster-Shafer evidence theory. Specifically, we incorporate a Causal Constraint Mechanism to mitigate spurious correlations through sample reweighting, thereby promoting the learning of domain-invariant morphological features. Furthermore, we design an Evidential Boundary Constraint module to model pixel-level uncertainty and suppress unreliable boundary responses in ambiguous regions, enabling more accurate boundary delineation. Extensive experiments on six datasets, including four public benchmarks and two in-house clinical datasets, demonstrate that CaEG-Net achieves superior segmentation performance in comparison to existing methods. The code was released at https://github.com/Bonjour-492/CaEG-Net.

Digital pathology has fundamentally altered diagnostic workflows by enabling the computational analysis of gigapixel Whole Slide Images (WSIs), yet effectively deciphering their complex tumor microenvironments remains a formidable challenge. Existing Multiple Instance Learning (MIL) frameworks typically treat Whole Slide Images as unstructured bags of patches, discarding critical morphological semantics and spatial geometry. This lack of inductive bias often leads to overfitting on background noise and fails to align visual features with high-level diagnostic knowledge. To overcome these limitations, we propose the Hierarchical Prototype-based Domain Priors (HPDP) framework, a unified multimodal approach applicable to diagnostic and prognostic tasks in histopathology. HPDP enhances conventional MIL models by introducing a Morphologically Anchored Prototype System (MAPS), which uses reference centroids derived from clustering training patch features to regularize learnable Prior Experts, together with a Sinusoidal Positional Encoder (SPE) to incorporate two-dimensional absolute positional information. Furthermore, a Hierarchical Cross-Modal Alignment (HCMA) module incorporates Large Language Model (LLM)-generated auxiliary descriptions to modulate visual representations. Experiments across seven cancer cohorts demonstrate competitive classification and survival prediction performance, with the highest mean classification metrics among the evaluated methods and cohort-dependent performance in survival analysis.

Introduction and aims: Head and neck squamous cell carcinoma (HNSCC) is a major oral health challenge, and accumulating evidence suggests that microbial dysbiosis may contribute to disease development. However, most previous studies relied on 16S rRNA sequencing, limiting species-level resolution and providing inconsistent microbial signatures across cohorts. This study aimed to characterize HNSCC-associated microbial alterations using genome-resolved metagenomics and identify candidate microbiota-based interventions. Methods: Shotgun metagenomic data from 150 oral and faecal samples spanning an in-house case–control cohort and two public datasets were integrated and analysed using a unified bioinformatics pipeline. Microbial diversity, differential biomarkers, co-occurrence networks, and species-level genome bins were analysed. Candidate probiotic strains were predicted using metaProbiotics and further evaluated in a CAL27 zebrafish xenograft model. Results: Faecal communities showed stronger disease-associated restructuring than oral communities, with reduced observed richness and significant Bray–Curtis separation, whereas oral community differences were not significant. Integration of LEfSe and random forest identified eleven faecal consensus markers, including control-enriched Roseburia faecis, Phocaeicola vulgatus , and Bifidobacterium longum , and four cancer-enriched oral markers, including reproducible Capnocytophaga sputigena . Microbial networks were consistently less connected in HNSCC across both niches. Among 990 dereplicated species-level genome bins, Bifidobacterium represented the dominant high-confidence probiotic genus. In vivo validation showed that B. longum and the induction chemotherapy component docetaxel, cisplatin, and 5-fluorouracil (TPF) reduced tumour burden and inflammatory/angiogenic responses, while combination treatment modulated macrophage-associated migration and partially restored microbial network connectivity. Conclusion: Genome-resolved metagenomic profiling revealed distinct oral and gut microbial features associated with HNSCC and identified B. longum as a candidate adjunctive probiotic intervention requiring further mammalian and clinical validation. Clinical relevance: This study provides microbial biomarkers and candidate probiotic strategies that may support future microbiome-based risk assessment and adjunctive management of HNSCC in oral healthcare settings.

Introduction and aims Head and neck squamous cell carcinoma (HNSCC) is a major oral health challenge, and accumulating evidence suggests that microbial dysbiosis may contribute to disease development. However, most previous studies relied on 16S rRNA sequencing, limiting species-level resolution and providing inconsistent microbial signatures across cohorts. This study aimed to characterize HNSCC-associated microbial alterations using genome-resolved metagenomics and identify candidate microbiota-based interventions. Methods Shotgun metagenomic data from 150 oral and faecal samples spanning an in-house case–control cohort and two public datasets were integrated and analysed using a unified bioinformatics pipeline. Microbial diversity, differential biomarkers, co-occurrence networks, and species-level genome bins were analysed. Candidate probiotic strains were predicted using metaProbiotics and further evaluated in a CAL27 zebrafish xenograft model. Results Faecal communities showed stronger disease-associated restructuring than oral communities, with reduced observed richness and significant Bray–Curtis separation, whereas oral community differences were not significant. Integration of LEfSe and random forest identified eleven faecal consensus markers, including control-enriched Roseburia faecis, Phocaeicola vulgatus, and Bifidobacterium longum, and four cancer-enriched oral markers, including reproducible Capnocytophaga sputigena. Microbial networks were consistently less connected in HNSCC across both niches. Among 990 dereplicated species-level genome bins, Bifidobacterium represented the dominant high-confidence probiotic genus. In vivo validation showed that B. longum and the induction chemotherapy component docetaxel, cisplatin, and 5-fluorouracil (TPF) reduced tumour burden and inflammatory/angiogenic responses, while combination treatment modulated macrophage-associated migration and partially restored microbial network connectivity. Conclusion Genome-resolved metagenomic profiling revealed distinct oral and gut microbial features associated with HNSCC and identified B. longum as a candidate adjunctive probiotic intervention requiring further mammalian and clinical validation. Clinical relevance This study provides microbial biomarkers and candidate probiotic strategies that may support future microbiome-based risk assessment and adjunctive management of HNSCC in oral healthcare settings.

Blueberries are highly susceptible to postharvest softening, oxidative damage, and microbial decay. In this study, gelatin/Sargassum fusiforme polysaccharide/carbon dots (gelatin/SFP/CDs) composite coating was applied to blueberries, which were then stored at 25 °C and 50–55% relative humidity for 12 days. The preservation mechanism was investigated from quality, physiological, and transcriptional perspectives. During 12-day storage, gelatin/SFP/CDs-coated blueberries exhibited lower decay incidence, darkening, weight loss, pH increase, and bacterial proliferation, while maintaining higher firmness, soluble solids content, titratable acidity, and commercially acceptable rate. Physiological analyses showed that gelatin/SFP/CDs coating delayed cell wall degradation by preserving cell wall polysaccharides and suppressing enzymes involved in cell wall degradation. Meanwhile, it alleviated oxidative stress by reducing reactive oxygen species (ROS) accumulation and enhancing antioxidant capacity. Transcriptomic analysis further revealed that gelatin/SFP/CDs regulated genes involved in cell wall metabolism, ROS generation, antioxidant defense, and the ascorbate–glutathione cycle. These findings suggest that gelatin/SFP/CDs coating delays blueberry senescence through suppressing cell wall degradation and alleviating ROS accumulation.