Browse Articles

Application of multiscale token fusion and pruning in CNN–Transformer hybrids under low-data training for image recognition

Abstract In this paper, a hybrid convolutional neural network (CNN) and transformer architecture for image classification that explicitly exploits multiscale spatial representations while maintaining computational efficiency is proposed. A convolutional backbone is first used to extract hierarchical feature maps, which are subsequently tokenized and fused through a cross-scale token fusion (CSTF) mechanism. In addition, several token-level and CNN-level pruning strategies are evaluated to examine whether redundant spatial tokens or convolutional features can be removed without substantially degrading performance. Extensive experiments on Caltech-101 and Oxford-IIIT Pets under low-data training conditions show that the best proposed configurations are dataset-dependent: the single 14 $$\:\times\:14$$ scale achieves 70.28 $$\:\pm\:1.91$$ % accuracy on Caltech-101, while the single 7 $$\:\times\:$$ 7 scale achieves 28.35 $$\:\pm\:1.62$$ % accuracy on Oxford-IIIT Pets. Among the multi-scale fusion models, 7 $$\:\times\:$$ 7+14 $$\:\times\:$$ 14 performs best on Caltech-101 69.68 $$\:\pm\:$$ 1.77%, whereas 7 $$\:\times\:$$ 7+14 $$\:\times\:$$ 14+28 $$\:\times\:$$ 28 performs best on Oxford-IIIT Pets 27.57 $$\:\pm\:$$ 1.05%. CNN kernel pruning achieves the strongest pruning performance on both datasets, reaching 68.37 $$\:\pm\:$$ 1.06% on Caltech-101 and 28.15 $$\:\pm\:$$ 1.15% on Oxford-IIIT Pets. These results indicate that multi-scale token fusion can provide competitive performance, but careful scale selection is more important than simply increasing the number of spatial token scales.

Inflammatory biomarkers of asymptomatic and symptomatic tuberculosis

Abstract A large proportion of individuals with tuberculosis (TB) are asymptomatic. The biological and inflammatory underpinnings of asymptomatic TB are unknown and may differ from symptomatic TB. We characterise blood transcriptomic and proteomic profiles in South African community screening vs. health facility-based triage cohorts. Asymptomatic TB shares core transcriptomic and proteomic features with symptomatic TB, including upregulation of innate, interferon and inflammatory pathways and downregulation of T and B cell pathways. Integration of transcriptomic and proteomic data from asymptomatic TB individuals identifies two distinct sub-clusters characterized by higher or lower bacterial burden, blood IFN-γ responses, BMI, and chest radiographic abnormalities, suggesting different disease severity. We identify a new blood transcriptomic signature of asymptomatic TB. However, diagnostic performance of transcriptomic and proteomic markers is weaker for asymptomatic TB than symptomatic TB, suggesting that policy development for community-based, asymptomatic TB screening should not adopt biomarkers developed for symptomatic TB triage without further optimization.

Intravenous hydrogen-rich acetated Ringer’s solution attenuates liver injury in heatstroke rats through endothelial glycocalyx preservation

Abstract Heatstroke is a life-threatening systemic disorder characterized by hyperthermia, inflammatory activation, endothelial injury, and multiple organ dysfunction. The liver is particularly vulnerable, but effective post-onset therapeutic strategies remain limited. This study examined whether intravenous hydrogen-rich acetated Ringer’s solution (H-AR) attenuates liver injury after heatstroke onset in rats, in comparison with acetated Ringer’s solution alone (Control) and 2% hydrogen inhalation (HI). Liver injury was assessed 60 min after onset using histological scoring and liver enzyme measurements. Circulating markers of inflammation and endothelial injury, endothelial glycocalyx ultrastructure, and 24-h survival were also evaluated. Thermal exposure was comparable among the heatstroke groups. Both H-AR and HI significantly reduced histological liver injury scores compared with Control; H-AR showed significantly greater attenuation than HI and was the only intervention that significantly reduced alanine aminotransferase levels. Both hydrogen-based interventions suppressed high-mobility group box 1 and syndecan-1, preserved endothelial glycocalyx structure, and improved 24-h survival compared with Control. These findings suggest that post-onset intravenous H-AR attenuates heatstroke-associated liver injury, potentially through reduced inflammatory activation and endothelial glycocalyx preservation, and may offer advantages over hydrogen inhalation with respect to hepatoprotection.

Mechanosensory encoding of surface mechanics optimizes locomotion

Abstract Locomotion - whether walking, running, or crawling - depends on the precise coordination of forces between the body and its surroundings. Two critical factors in this process are the force that resists the relative motion between two bodies, and mechanosensation, the body’s ability to sense and respond to mechanical forces. Together, these processes enable efficient locomotion across diverse environments, yet how body-environment interaction forces shape neural activity in freely moving animals remains poorly understood. Here we show that the ‘gentle touch’ receptor neurons (TRNs) in the Caenorhabditis elegans body wall are sensitive to dynamic surface mechanics. Using a combination of calcium imaging and traction force microscopy in freely behaving animals informed by micromechanical stimulation, connectome-scale simulations, and optimal control theory, we identify friction force and crawling velocity as key determinants of mechanoreceptor activity. Mutations disrupting MEC-4 DEG/ENaC ion channel activity in body wall mechanoreceptors produce lethargic animals with impaired proprioceptive regulation, suggesting functional coupling between surface mechanoreceptors and proprioceptors. Together, our findings reveal that gentle ‘touch’ receptors encode body-environment interaction forces during movement to optimize locomotor energetics.

Integrated metabolomics and network pharmacology reveals that icariin antagonizes malignant phenotypes in triple-negative breast cancer via AKT1/SRC/NFKB1-mediated PI3K‒AKT pathway inhibition

Scaling and Population Loss in Mexican Urban Centres

Abstract Despite its pervasive implications, many cities worldwide continue to expand in a fragmented, horizontal manner. We analyse urban growth dynamics in 69 Mexican metropolitan areas from 1990 to 2020 using census data, developing a model of urban form change based on population size, density, and spatial configuration. We employ a radial probability density function and the urban expansion factor to create a framework for comparing urban expansion over time and across different regions. Over the past three decades, Mexico’s urban population has nearly doubled. However, populations have shifted outward, resulting in a decline of 2.5 million residents in central areas. Our analysis shows that distances from the city centre have increased by 28% on average, driven by population losses in central zones combined with growth in peripheral regions.

Multi-scale spatiotemporal analysis of landscape ecological responses to railway corridor development

Unidentifiability and false-positive inference in state-dependent diversification models

Behavioral evidence for visually dominant audiovisual motion integration

Biogeochemical signal from marine heatwaves, cold spells, and transient warming events in a coastal upwelling system

Spatiotemporal evolution and influencing factors of the coupling coordination between Agricultural Digitalization and Agricultural New Quality Productive Forces

Mechanistic study on the influence of organic matter on soil compressive strength and nonlinear resistivity

Antimicrobial peptide LL-37 promotes proliferation of human bone marrow-derived mesenchymal stem cells

Characterizing circular and linear non-coding RNA in human aged brains and their association with dementia

Long-term, large-current operation of superconducting power transmission that carried five million passengers

A forecast model for ionospheric TEC during solar flare events based on low-latitude GPS stations data

Effects of WeChat-based palliative care intervention on cancer-related fatigue among lung cancer patients: a randomized controlled trial

New dimensions in burnt bone taphonomy through X-ray, neutron imaging and multivariate analysis

E-commerce video & content understanding via vision: a comprehensive study on lip reading and visual speech recognition for enhanced shopping experience

Microstructure-driven enhancement of cement mortar using high-entropy oxides for improved strength, durability, and corrosion resistance

Abstract This study investigates the influence of high-entropy oxide (HEO) fine powder with submicron agglomerates (Mg₀.₂Ni₀.₂Co₀.₂Cu₀.₂Zn₀.₂O) on the mechanical performance, microstructural development, and durability characteristics of cement-based mortars. HEO was synthesized via a modified sol–gel method and incorporated into mortar mixtures as an additional additive at dosages of 0–1.0 wt% relative to the cement content. The experimental results demonstrate a strong dosage-dependent behavior. The optimum performance was obtained at 0.25 wt% HEO, where compressive and flexural strengths increased by up to approximately 3–4% compared to the control sample. In contrast, higher dosages (≥ 0.75 wt%) resulted in a gradual reduction in mechanical performance, indicating a threshold beyond which HEO begins to negatively affect hydration-related processes. Microstructural analysis (XRD and SEM–EDS) revealed that HEO particles contribute to matrix densification through a filler effect and act as nucleation sites for hydration products, leading to a more compact cementitious structure at optimal content. However, excessive incorporation appears to disrupt hydration continuity due to particle agglomeration and dilution effects. Durability tests showed improved resistance to water absorption, chloride ion penetration, and corrosion current with increasing HEO content, attributed to pore refinement and reduced ion transport pathways. The most balanced performance in both mechanical and durability properties was achieved at low HEO dosage. Overall, high-entropy oxides demonstrate significant potential as nano-scale additives for cementitious systems; however, their effectiveness is highly dependent on dosage, with low replacement levels providing the most beneficial performance balance.