CCTAD: A topologically associating domains detection method integrating convolutional autoencoder and hierarchical clustering.

Hydrogen-fueled internal combustion engines (H 2 -ICEs) hold strong potential as a pathway toward CO 2 -neutral propulsion. To reduce emissions, H 2 -ICEs are usually operated under fuel-lean conditions, where the flames are prone to thermo-diffusive instabilities (TDIs). These TDIs govern both local and global flame propagation, but their impact on full-scale engine combustion remains an open question. In this study, high-fidelity three-dimensional large-eddy simulations (LES) are performed at multiple mesh resolutions, with the finest grid sufficiently resolved to directly characterize flame front dynamics relevant to engine-scale combustion. The simulations reveal cellular and finger-like flame structures characteristic of TDIs throughout the entire combustion process. Analysis of the local thermo-chemical state demonstrates that differential diffusion induces pronounced mixture stratification and elevates reaction rates, resulting in super-adiabatic temperatures that strongly correlate with flame curvature. Building on these findings, the performance of the baseline artificially thickened flame (ATF) model and a recently developed thermo-diffusive (TD)-aware extension is assessed. Unlike the state-of-the-art ATF model, which suffers from grid dependence and underestimates the experimental pressure trace, the TD-aware formulation captures experimental trends more accurately and provides consistent, grid-independent integrated heat-release (IHR) traces. For the operating condition considered here, the results show that TD effects represent sub-grid-scale contributions that need to be accounted for to obtain consistent predictions of global combustion behavior under the investigated lean H 2 -ICE conditions. • LES captured cellular and finger-like thermo-diffusive flame structures. • Coarser grids suppressed fine-scale instabilities resolved at high resolution. • Local mixture stratification enhanced reactivity and caused super-adiabatic states. • ATF model showed grid bias from missing thermo-diffusive instability treatment. • Thermo-diffusive-aware ATF model reduced grid bias and improved predictive accuracy.

Comparative evaluation of deep learning architectures for bioclast classification in atomic force microscopy images

This paper proposes a deep learning-based segmentation model for biomimetic underwater acoustic communication systems, capable of extracting dolphin whistle signals from underwater acoustic data containing diverse noise sources while reducing contour discontinuity. The proposed model is based on a U-Net architecture and incorporates the Convolutional Block Attention Module (CBAM) and the Multi-Scale Combining Spatial Attention Module (MSC-SAM) to capture whistle features at multiple resolutions. CBAM is applied to skip connections to emphasize meaningful channels and spatial information related to whistle contours. In addition, MSC-SAM is designed to integrate spatial attention maps generated at each resolution, thereby reducing whistle contour discontinuity that occurs during the segmentation process. The proposed model is evaluated using acoustic data collected by the National Oceanic and Atmospheric Administration (NOAA) and compared with existing models, including U-Net, ResU-Net, U-Net++, FPN, PSPNet, DeepLabv3+, and MA-Net. Experimental results show that the proposed model achieves improved segmentation performance in terms of pixel accuracy, IoU, and Dice score. For the proposed whistle-level evaluation criteria, namely Discontinuous Whistles and Missed Whistles, the proposed model reduces discontinuous whistles and missed detections by 65.6% and 61.1%, respectively, relative to U-Net.

The clearance of apoptotic cells by macrophages, termed efferocytosis, reprograms macrophages to a resolution/repair phenotype, and pathologic defects in efferocytosis drive many chronic inflammatory diseases. Previous studies have elucidated numerous downstream pro-resolving pathways activated by efferocytosis, but whether there exists a common upstream trigger of these pathways remains unknown. Here, we report that efferocytosing macrophages surprisingly use a signaling module typically associated with inflammation to carry out this key initiating role in tissue resolution. The binding of apoptotic cells to the MerTK receptor triggers a rapid and transient activation of inhibitor of nuclear factor (NF) kappa-B kinase subunit beta (IKKβ), leading to NFκB and p38-signal transducer and activator of transcription 3 (STAT3) signaling and then activation of several key downstream pro-resolving pathways, including interleukin-10 (IL-10) production, continuing efferocytosis, and regulatory T (T reg ) cell expansion. The upstream IKKβ pathway and the downstream resolution pathways are linked through several intermediary molecules, including the transcription factor Myc, the epigenetic modifier ten-eleven translocation-2 (TET2), and the immune checkpoint protein programmed cell death ligand 1 (PD-L1). Deletion of macrophage IKKβ in vivo blocks the above resolution pathways and compromises tissue repair in two efferocytosis-mediated repair settings: resolution of thymic injury after dexamethasone-induced thymocyte apoptosis; and, most importantly, atherosclerosis regression induced by low-density lipoprotein (LDL)-lowering, which is highly relevant to the prevention of cardiovascular disease in humans. These findings illustrate the existence of a unifying upstream signal for efferocytosis-induced resolution, which could suggest new therapeutic strategies to enhance multiple tissue resolution pathways and to optimize anti-inflammatory therapies by avoiding blocking IKKβ-NFκB/p38-mediated resolution.

Background/Objectives: Cancer remains one of the leading causes of mortality worldwide, with lung and colon cancers among the most prevalent. Conventional histopathological diagnosis is time-consuming, requires expert pathologists, and is susceptible to human error. Methods: To address these limitations, this study proposes an automated classification framework for lung and colon cancer using histopathological images. The proposed method employs a lightweight pretrained deep learning model, MobileNetV3-Small, through transfer learning. Training is performed on an enhanced version of the LC25000 dataset, in which redundant image patches are removed to improve robustness and clinical generalizability. The images were initially available in multiple resolutions, which are resized to 224 × 224 × 3 to match the canonical input size of MobileNetV3-Small. Deep features are extracted from the dropout layer as it provides regularized representation of high-level features by reducing the overfitting (dimension N × 1024), which are optimized using a differential evolution algorithm, reducing the feature space to N × 60. These optimized features are evaluated using multiple classifiers. Results: Experimental results demonstrate a maximum classification accuracy of 98.14% using a Quadratic Support Vector Machine (SVM) and a 21.3× speed-up achieved with bagged trees, outperforming several state-of-the-art approaches representing a 3.34% improvement over the baseline study on the enhanced dataset. Conclusions: The results confirm that the proposed framework effectively balances high accuracy with computational efficiency. The use of a lightweight deep model combined with feature optimization makes the approach well-suited for practical clinical environments.

Haul roads in surface mining are critical infrastructure directly influencing operational productivity, safety, and costs. However, these networks change frequently due to ongoing mining activities, making traditional mapping methods impractical for large-scale or rapidly evolving sites. Remote sensing imagery offers a scalable alternative, yet complex backgrounds, variable road widths, and spectral similarities between roads and surrounding surfaces make accurate extraction challenging. This study proposes HRM-Net, a hybrid transformer–CNN autoencoder framework for automated extraction of mine haul roads from remote sensing imagery. HRM-Net introduces inception-like patch embedding to capture local contextual information and employs a manifold-constrained hyper-connection strategy in the attention and fusion blocks to enhance information flow across the architecture. This hierarchical design enables progressive learning of discriminative semantic representations across multiple spatial resolutions, critical for road extraction in cluttered mining environments. Trained and evaluated on diverse mine sites, HRM-Net achieved 92.53% overall accuracy, 85.12% F1-score, 75.57% mIoU, 83.57% precision, and 86.94% recall, outperforming state-of-the-art transformer-based and CNN-based segmentation models. Furthermore, model interpretability was analyzed through linear probing and boundary alignment evaluations. Results demonstrate that discriminative features emerge at early network stages and are effectively preserved throughout the architecture, while boundary predictions exhibit superior consistency compared to existing approaches.

Epileptic seizure detection from Electroencephalography (EEG) signals is challenging due to their complex temporal dynamics and intricate inter-channel dependencies. A key difficulty lies in capturing multi-scale temporal features that span different time ranges. Seizure-related EEG patterns include rapidly varying micro-scale features and longer-duration macro-scale features, and modeling them jointly often leads to feature interference, hindering accurate temporal representation. In addition, many existing approaches fail to effectively capture spatial relationships between brain regions, further limiting detection performance. To cope with issues above, we design the Adaptive Multi-scale Spatiotemporal Mixing Network (AMSMN). The framework first decomposes EEG signals into macro- and micro-scale sequences, which are processed independently across multiple temporal resolutions to reduce cross-scale interference and better represent temporal dynamics. A spatial attention mechanism then fuses the decomposed features, ensuring that important inter-channel information is preserved. Finally, an Informer-based sparse attention layer captures long-range dependencies, allowing the model to focus on the most relevant global interactions across brain regions. We carry out experiments on two available databases demonstrate that AMSMN consistently achieves superior performance to prior methods in both patient-specific and cross-patient settings. The results confirm that the proposed framework improves seizure detection accuracy, robustness, and generalization by effectively integrating multi-scale temporal modeling with global spatial dependency extraction. This work advances EEG-based seizure detection by enabling precise multi-scale temporal analysis and efficient global dependency modeling, offering strong performance and generalizability for clinical applications.

The rapid expansion of large-scale image collections demands content-based image retrieval systems that are both semantically reliable and computationally efficient. Conventional CBIR approaches based on isolated visual descriptors often fail under high intra-class variation and inter-class ambiguity. This work proposes a causality-enhanced multiresolution residual learning framework that integrates deep representation learning, causal modeling, and lightweight optimization for robust image retrieval. A multiscale ResNet 50 backbone captures complementary fine-grained and high-level semantic features across multiple resolutions, while a causal variational autoencoder explicitly models latent spatial dependencies to improve semantic consistency and interpretability. Feature redundancy is minimized by an enhanced fast osprey optimization algorithm, enabling compact, discriminative feature selection at low computational cost. Extensive experiments on CIFAR-10, Oxford Flowers, and Corel 1000, using fivefold cross-validation, demonstrate consistent improvements over recently published thirteen state-of-the-art methods. Further gains in mean average precision, normalized discounted cumulative gain, and reduced retrieval time confirm the effectiveness, robustness, and practical viability of our approach across diverse retrieval scenarios. Moreover, the retrieval samples exhibit strong robustness against high intra-class variability and severe inter-class ambiguity, preserving discriminative consistency across visually similar categories.

Accurate models of species distributions in mountainous ecosystems are essential for effective conservation, yet reliance on open‑access medium‑resolution elevation products often obscures fine-scale habitat heterogeneity critical for wildlife. We evaluated whether high-resolution digital surface models (DSMs) (≤1m) based on data from an unmanned aerial vehicle (UAV) enhanced species distribution models (SDMs) relative to medium‑resolution satellite-based ALOS World 3D (AW3D) DSMs (≥30m) for François' langur (Trachypithecus francoisi). This endangered primate represents an ideal focal species for evaluating SDM performance because its habitat consists of topographically complex karst mountains. Using a systematic comparative framework, we built 290 random forest SDMs across multiple spatial resolutions (3-150m) and sample sizes (30-122 occurrence records). We assessed model accuracy with the area under the receiver operating characteristic curve (AUC-ROC) and the area under the precision-recall curve (AUC-PR). The UAV DSM improved model performance across all metrics, particularly identification of microhabitat. Performance improvements were scale dependent and most pronounced at fine spatial resolutions of 3-10m. Improvement was less at coarser resolutions above 100m, where topographic aggregation reduced fine-feature discrimination. The UAV-based models were more accurate than AW3D-based models even with fewer presence records, indicating greater robustness in data-limited mountainous surveys. Slope-related predictors derived from UAV DSM were more important than those derived from AW3D DSM, suggesting that high-resolution elevation data more effectively resolve terrain features relevant to species habitat selection. The UAV DSM showed wildlife corridors in steep terrain that did not appear in the AW3D DSM. Given that contemporary conservation planning in mountainous protected areas predominantly relies on medium-resolution elevation data, we recommend implementing UAV structure from motion photogrammetry as a cost-effective approach for generating detailed elevation data to enhance biodiversity management in topographically complex landscapes.

BackgroundRadiomics has emerged as a promising approach for extracting quantitative features from medical images to support tumor characterization and treatment response assessment. With the increasing use of magnetic resonance-guided linear accelerator (MR-Linac), ensuring the stability and reproducibility of radiomics features derived from magnetic resonance imaging (MRI) has become critical for reliable clinical applications. However, image preprocessing parameters may substantially influence feature stability. Therefore, this study aimed to evaluate the effects of image preprocessing parameters on radiomics feature stability in MRI acquired on a 1.5T MR-Linac system, with specific assessment of test-retest repeatability, longitudinal repeatability, and inter-platform reproducibility.MethodsMRI datasets were acquired using the American College of Radiology (ACR) phantom on 1.5T MR-Linac systems for test-retest, longitudinal, and inter-platform analysis. The T1-weighted (T1w), T2-weighted (T2w), and fluid-attenuated inversion recovery (FLAIR) sequences were collected. Five regions of interest were delineated on T1w images and propagated to corresponding T2w and FLAIR sequences. Image preprocessing strategies included voxel resampling (multiple isotropic resolutions), intensity normalization (none or Z-score), and intensity discretization using bin width (BW) or bin number (BN). Feature stability was assessed using the intraclass correlation coefficient (ICC) and coefficient of variation (CV). Features with ICC values >0.9 and CV values <10% were considered robust.ResultsOptimal preprocessing strategies varied across imaging sequences and evaluation tasks. In the test-retest analysis, the proportion of robust features reached 85.87% for T1w, 89.13% for T2w, and 90.22% for FLAIR sequences under optimal settings. In contrast, longitudinal repeatability and inter-platform reproducibility showed substantially lower stability, with robust feature proportions ranging from 42.39% to 76.09% across sequence and preprocessing configurations. Larger voxel sizes (>2 mm isotropic) consistently reduced stability across all tasks. The BN discretization method generally yielded higher proportions of robust features than the BW method; however, this advantage was sequence- and task-dependent. Z-score normalization had minimal effect when applied with BN discretization, but reduced feature stability when combined with the BW discretization.ConclusionsImage preprocessing parameters significantly influence the stability of radiomics features acquired on MR-Linac. Stability varies considerably between test-retest repeatability and longitudinal repeatability or inter-platform reproducibility. Task- and sequence-specific optimization of preprocessing strategies is therefore essential before clinical implementation. Further validation in clinical datasets is acquired to support robust integration of MR-Linac radiomics into adaptive radiotherapy workflows.

Corn (Zea Mays L.) faces challenges from leaf diseases, which become severe when farmers lack the expertise to recognize and manage them. This study presents a comparative analysis of VGG16 and VGG19 architectures for detecting corn leaf diseases, highlighting their performance under standardized conditions using transfer learning. The novelty of this study lies in the direct benchmarking of both models across multiple image resolutions and training epochs, which has not been comprehensively explored in previous studies. The system categorizes diseases based on images, thereby helping farmers manage corn leaf diseases more effectively. The VGG16 architecture was chosen for its balance of depth and computational efficiency, while VGG19 offers higher accuracy due to its increased layer depth and complexity. This system is expected to assist farmers in detecting corn leaf diseases more efficiently and accurately than previously possible. The dataset used in this study consists of 4198 images, divided into four categories: Healthy, Blight, Common Rust, and Gray Leaf Spot. The dataset was split into 80% for training and 20% for testing purposes. The classification results using 2 architectures, VGG16 and VGG19, with the use of the SGD optimiser, show that VGG19 outperforms VGG16. The VGG19 model demonstrated a performance level of 92.74% accuracy, alongside 91% for precision, recall, and F1-score. In comparison, VGG16 achieved a slightly lower accuracy of 92.62%, with precision at 91%, recall at 89%, and an F1-score of 90%. This performance variance is attributed to the architectural depth, as VGG19 utilizes 19 layers while VGG16 is limited to 16. Ultimately, this tool aims to provide farmers with a more precise and streamlined method for identifying corn foliage conditions.

The cerebral vasculature is central to brain function, with alterations linked to numerous cerebrovascular and neurological disorders. Yet, no single imaging modality can capture the entire cerebral vascular network in humans. Instead, an array of techniques are sensitized to different spatial scales, while trading off resolution for coverage. Magnetic Resonance Imaging (MRI) typically resolves only large pial vessels, while high-resolution microscopy allows micrometer-scale vessels to be mapped over limited spatial extents. These techniques must therefore be combined to obtain a complete mapping of the cerebral angioarchitecture, which underscores the need for automatic, cross-modal vessel segmentation. Here, we introduce VesSynth, a flexible vessel segmentation framework that achieves state-of-the-art accuracy across multiple modalities and spatial resolutions (MR, optical and X-ray imaging), despite being trained entirely on synthetic data. By enabling consistent vascular mapping across scales, this framework paves the way to comprehensive investigation of cerebrovascular organization and its role in health and disease.

A multiscale convolutional autoencoder for fusing element concentrations and key geological information for geochemical anomaly recognition

Abstract Tropical cyclones are among the most destructive weather systems, yet prediction skill in the South-west Indian Ocean (SWIO) lags due to limited observations and modelling constraints. This study evaluates the Conformal-Cubic Atmospheric Model (CCAM) in simulating track and intensity of Tropical Cyclone Idai, relative to the operational Unified Model (UM) at the South African Weather Service (SAWS). Simulations were performed at multiple resolutions (CCAM: 25 km, 6 km; UMGA: 10 km; UM: 4.4 km) and lead times (72 h, 48 h, 24 h), validated against independent observational and reanalysis datasets. Both models reproduced Idai’s track more accurately at finer resolution and shorter lead times. CCAM exhibited a southward track bias, while UM deviated northward. Landfall position and timing were generally captured, with position errors reduced to &lt; 30 km at 24 h, although CCAM 6 km showed a late landfall bias of ~3 h at 72 h. Intensity was underestimated by all simulations, however, CCAM 6 km better matched best-track winds and central pressures, while UM 4.4 km aligned more closely with ERA5. Rainfall patterns differed, with CCAM overestimating rainfall extent and intensity, whereas UM 4.4 km more realistically captured IMERG and CHIRPS patterns, though with weaker magnitudes. Results show that high horizontal resolution is crucial for representing TC characteristics, with microphysical processes becoming increasingly influential at convection-permitting scales. Differences in model initialisation also contributed to track and intensity biases. These findings underscore the value of high-resolution modelling, improved initialisation and observations to advance TC forecasting and early warning in the SWIO.

Abstract We present the second release of the GR-Athena++ waveform catalog, comprising four new quasi-circular, non-precessing, spinning binary black hole simulations. These simulations are performed at high resolutions and represent a step toward generating high-fidelity gravitational waveforms that can eventually meet the accuracy requirements of upcoming next-generation detectors, including LISA, Cosmic Explorer, and Einstein Telescope. Gravitational waves are extracted at future null infinity using both Cauchy characteristic extraction and finite-radius extraction. For each simulation, we provide strain data across multiple resolutions and analyze waveform accuracy via convergence studies and self-mismatch analyses. The absolute phase and relative amplitude differences reach their largest values near the merger, while the smallest errors are of order O(10 -2 ) and O(10 -3 ), respectively. A self-mismatch analysis of the dominant (2, 2) mode yields mismatches between O(10 -5 ) and O(10 -7 ) for a total binary mass of 10 6 M ⊙ over the frequency range 0.002 to 0.1 Hz using LISA's noise curve. All waveforms are publicly available via ScholarSphere.

We present the most comprehensive sampling of the macaque and human brain with diffusion MRI to date. As part of the BRAIN CONNECTS center for Large-scale Imaging of Neural Circuits, we leverage ultra-high-gradient MRI systems, including the first-of-its-kind Connectome 2.0, for post-mortem acquisitions. Each sample is imaged for ~250 hours at multiple spatial resolutions down to 0.25 mm isotropic for whole macaque brains and 0.4 mm isotropic for human hemispheres. Our optimized protocols allow us to sample both species across ~50 diffusion shells varying in b-value, diffusion time, and echo time, reaching ultra-high b-values up to 64000 s/mm^2 with high signal-to-noise ratio. We demonstrate that these multi-dimensional data resolve not only white matter connectional architecture but also cortical and subcortical cytoarchitectonic boundaries, at a level of detail previously inaccessible in whole-brain noninvasive imaging. As such, these data are an important resource for both technical development and basic and clinical neuroscience.

Structural colouration arises from the interaction of light with nanoscale structures and offers sustainable alternatives to pigment-based colours. However, current structural colour patterning methods rely on multi-step lithographic processes or multiple ink formulations, limiting scalability and spatial resolution. Inspired by melanosome self-assembly in bird feathers, we develop a one-step, mask-free strategy to generate high-resolution structural colour patterns via tunable nanoparticle segregation. During photocuring, silica nanoparticles dispersed in acrylic resin migrate toward oxygen-permeable substrates, forming a nanoparticle-enriched disordered layer. Such segregation is driven by interfacial oxygen inhibition and kinetically governed by the photocuring rate. Using grayscale digital light processing printing, we programmably control the local segregation thickness to create high-resolution structural colour patterns for visual display and information encryption. The segregation structure also affects mid-infrared reflectivity, allowing for infrared camouflage. This scalable approach establishes a mechanistically guided route to multifunctional photonic materials.

Accurate diagnosis of clinically significant prostate cancer (csPCa) from multi-view MRI scans (axial, sagittal, and coronal) is essential for effective treatment planning and improved outcomes. Although deep learning has advanced prostate MRI analysis, many existing approaches adopt late fusion strategies that aggregate one-dimensional feature vectors extracted independently from each view, resulting in loss of spatial information and anatomical correspondence across views, ultimately limiting diagnostic performance. While Vision Transformers offer flexibility in processing multi-view patches, their memory requirements scale quadratically with the number of patches, hindering efficient concurrent processing. In contrast, Swin Transformers efficiently capture local features but are typically restricted to single-view processing due to their reliance on regular-grid input constraints. To overcome these limitations, we propose a hybrid fusion framework that decomposes multi-view information integration into iterative intra-view and inter-view interactions across multiple resolutions. The framework preserves spatial coherence and enables fine-grained feature integration while maintaining computational efficiency. Specifically, the inter-view feature exchange module, based on the Vision Transformer, employs bridge tokens to summarize information from localized patch windows, reducing memory usage while preserving spatial relationships across views. The intra-view feature extraction module, built on the Swin Transformer, facilitates dynamic, attention-driven interactions among image patches and bridge tokens within each window. Moreover, shared positional embeddings are explicitly incorporated to enhance spatial correspondence across views. Extensive experiments on a public dataset demonstrate the superiority of our method in csPCa classification. Ablation studies highlight contributions of different components, while attention map visualizations validate integration of anatomical structures across views.

This work presents a numerical-experimental validation of aeroacoustic predictions for bio-inspired leading edge serrated blade cascades. Transient simulations were carried out on a four-blade cascade using several turbulence modeling strategies commonly applied in broadband noise analysis-Spalart-Allmaras (SA), k-ω SST, k-ε, Scale-Adaptive Simulation (SAS), and Large Eddy Simulation (LES)-for assessing their capability to reproduce measured spectra. Multiple timestep resolutions were tested to ensure temporal accuracy. The comparison indicates that below 900 Hz, interaction noise is difficult to evaluate for such applications, whereas in the range from 0.9 to 5 kHz the turbulent jet-blade interaction is clearly captured. In the low-frequency regime (<1 kHz), the SA, SAS, and k-ω SST models exhibit similar behavior, while at higher frequencies SAS provides the closest agreement with experimental results, albeit with a slight tendency to overestimate at the upper end of the spectrum. LES demonstrates a satisfactory performance in reproducing the baseline response. The validation of numerical simulations with experimental results has been achieved, and a complex analysis using pressure measurements on the blade surface for a four-blade cascade configuration shows that turbulent formations lose their coherence quite significantly across several frequency bands. Overall, the results confirm that numerical simulations can reproduce the dominant experimental trends, while emphasizing the model-dependent trade-offs in predicting the acoustic benefits of bio-inspired leading edge serrations.