Patch Size Research Articles

Ageing and environment affect female more than male color in a mutually ornamented bird

Abstract Color ornamentation is often sexually selected as a signal of condition, which may be affected by the environment and senescence, but long-term studies in mutually ornamented species are rare, limiting our understanding of sex-specific senescence. We measured color ornamentation of 60 common waxbills (Estrilda astrild) after capture from the wild and in 4 autumns in a large open-air mesocosm. We found longitudinal decreases in the size of the red mask and redness of plumage and bill color, consistent with senescence. Bill color decreased more in females. In males, there was a longitudinal increase in the size of the red breast patch. We also found increased color ornamentation, especially in females, when moving from the wild to the mesocosm, likely due to environmental effects. Monitoring waxbills in bird cages suggested that bill and plumage redness peak in the breeding season. Stronger senescence and environmental effects in females indicate cautious investment in ornamentation, while weaker senescence in males could be due to stronger sexual selection. Overall, although similarly ornamented, the sexes differ in their temporal investment in ornamentation, in line with their life-histories.

Enhancing Real-Time Vision Systems: Integrating Dynamic Vision Sensors with Vision Transformers to Increase Computational Efficiency

Abstract. Optimizing and understanding vision systems' computational accuracy is crucial as they become increasingly integrated into daily life. Dynamic Vision Sensors (DVS) and Vision Transformers (ViTs) lead computer vision technology with efficient object recognition and image processing. However, DVS data's high computational complexity poses a problem for its real-time implementations. Merging these technologies can enhance vision system performance in dynamic environments and optimize real-time DVS processing. In our work, we use a ViT architecture to classify the DVS 128 dataset and compare our results with existing works using SNNs. We analyze how our method affects accuracy and loss, experimenting with different DVS-to-ViT input patch sizes. Our results show that the large patch size of 32x32 pixels has better accuracy and smaller loss than the 4x4 pixel patches as epochs increase. Our method also achieved a high 98.4% accuracy and low 0.22 loss within five epochs, significantly outperforming previous works averaging 93.13% accuracy over more epochs. These results highlight ViTs' large potential for real-time DVS data classification in applications that require high accuracy, like autonomous vehicles and surveillance systems.

SEISMIC WAVE PROPAGATION IN PARTIALLY SATURATED FRACTAL POROUS MEDIA

Wave-induced fluid flow, being widely accepted as a dominant loss mechanism of wave attenuation, can lead to significant seismic attenuation due to mesoscopic heterogeneities such as partial saturation. However, the dependence of fluid distribution on microstructure in partially saturated porous media remains unclear. To quantify the relationship between microstructure and fluid distribution, a saturation fractal dimension is applied on the assumption of pore fractal distribution and gas patch fractal distribution. By integrating the Biot-Rayleigh theory, a theoretical model of partially saturated fractal porous media is established. The results indicate that the scales of fluid flow and the magnitude of peak attenuation are significantly influenced by both the maximum gas patch size and the pore fractal dimension. The proposed model is further validated by comparing it with laboratory measurements. The findings indicate that the variations in the maximum gas patch size correspond to the variations in fluid distribution, and the peak attenuation will reach its maximum magnitude when the saturation fractal dimension equals the pore fractal dimension. The discrepancies between the measurements and modeling results are discussed, revealing that, in addition to microstructure, factors such as rock properties, boundary condition, and saturation methods significantly influence the fluid distribution, as well as velocity dispersion and attenuation. The proposed theory provides a reasonable explanation of dispersion and attenuation in fieldwork, thereby presenting a novel perspective for future endeavors in forward modeling and seismic inversion.

Anthropogenic land consolidation intensifies zoonotic host diversity loss and disease transmission in human habitats.

Anthropogenic land-use change is an important driver of global biodiversity loss and threatens public health through biological interactions. Understanding these landscape-ecological effects at local scales will help achieve the United Nations Sustainable Development Goals by balancing urbanization, biodiversity and the spread of infectious diseases. Here, we address this knowledge gap by analysing a 43-year-long monthly dataset (1980-2022) of synanthropic rodents in Central China during intensive land-use change. We observed a notable increase in the mean patch size, coinciding with a substantial change in rodent community composition and a marked decline in rodent diversity; eight of the nine local rodent species experienced near-extirpation. Our analysis reveals that these irregular species replacements can be attributed to the effect of land consolidation on species competition among rodents, favouring striped field mice, a critical reservoir host of Hantaan virus (HTNV). Consequently, land consolidation has facilitated the proliferation of striped field mice and increased the prevalence of HTNV among them. This study highlights the importance of considering both direct and indirect effects of anthropogenic activities in the management of biodiversity and public health.

Circularly Polarized Variations of Truncated Corner Square Microstrip Antennas Employing a U-slot

Circularly polarized designs of square microstrip antennas employing U-slot are presented in a 2400 MHz frequency spectrum. Design with a U-slot employing unequal lengths or the corner truncated patch with equal length U-slot yields axial ratio bandwidths of 3.84% and 2.2%, respectively. Against the conventional offset feed patches or the truncated corner square patches, the proposed designs decrease the frequency and patch size by 10% and 22%, respectively. Reduction in the frequency and patch area, while employing the U-slot is not being discussed in the reported papers on the circularly polarized designs, and hence it is the new contribution in the present study. The U-slot cut configurations exhibit a radiation pattern in the broadside with a gain of larger than 6 dBi across the complete reflection coefficient bandwidth. To obtain a wideband and circular polarized response using a single patch, a reconfigurable design of a corner truncated square patch employed with a U-slot is presented. A design methodology is presented based on the resonant length formulation to obtain reconfigurable U-slot cut design as per specific band start frequency in wideband response or center frequency in the axial ratio bandwidth in the circular polarized antenna. This helps to design these configurations for the specific frequency band. Simulated results in the proposed design are verified using the measurements that reflect a close agreement.

Pre-vascularisation of the acellular bladder matrix using the omentum in a porcine model prior for bladder reconstruction-a step towards clinical application?

Research studies with porcine acellular bladder matrix (PABM) showed integration of only small sized stamps in recipient bladders, however for clinical use in bladder augmentation significantly larger patches are needed. We hypothesised pre-vascularisation with omentum may be a step towards clinical translation. Eight domestic pigs were operated three times 8-10weeks apart: 1-Implantation; PABM with recorded dimensions were sutured around a tissue expanding device, wrapped in omentum and sutured to the anterior abdominal wall. 2-Augmentation; hemi-cystectomy and bladder augmentation was performed with the pre-vascularized PABM using non-absorbable suture 3-Sacrifice; The dimensions of the PABMs were measured macroscopically, the in-vivo microcirculation of the PABMs were assessed using laser speckle contrast imaging. HE staining, uroplakin 3 and CK7 immunohistochemistry was performed. In seven animals, the bladder augmentation was successful without complication. One animal was lost in bowel obstruction and in two animals enteric fistula was found after the first intervention. The rectangular shape of the initial tissue expander was subsequently changed. All the seven patches were strong, compliant and had integrated with the surrounding native bladder and were 83% (48-100%) of the original patch size. Laser speckle contrast imaging already showed microcirculation at the patch edges at augmentation and this further improved towards the centre of the patches by the end of the study. Histology demonstrated acute inflammatory response with fibroblast infiltration after implantation and full coverage by urothelium was seen with positive staining for CK7 antibodies. Pre-vascularization of PABM in the omentum of healthy porcine models allows larger PABM patches to integrate this may be a step towards clinical application.

Why there are more species in several small patches versus few large patches: A multispecies modelling approach

Abstract As global biodiversity is rapidly declining due to habitat loss, it is important to determine how to protect it. The policies of many conservation agencies are still prioritizing the preservation of a single large habitat area (SL) versus several small areas of the same size (SS), despite empirical support favouring SS. However, to date, while many studies have explored this issue using model communities, the results are mixed. In competitive community models, fragmentation has a negative effect at low habitat amounts, whereas neutral community models suggest positive effects. This work aims to explore spatial multispecies models to verify the SS > SL pattern and determine its causes. We used three types of models: a classical neutral, neutral with habitat selection and a competitive hierarchical model. All models have three parameters, namely extinction, reproduction and dispersal distance. In the classical neutral model, species are equivalent and have the same parameters; when there is a reproduction event, they send the propagule to a surrounding patch blindly, and if the patch is already occupied, the propagule dies. In the neutral with habitat selection model, the propagules select an empty patch, so their survival is ensured. In the hierarchical model, species have competitive dominance and more dominant ones can replace the less dominant with a replacement rate parameter. We use 100 species and start simulating the colonization of an empty area; then, we destroy a fixed amount of habitat using different regular patch sizes. The results for all models are almost always that SS retains more species than SL. The extinction of species in patches depends on the quotient between reproduction and extinction rates in relation to patch size. The only case when SL > SS is when dispersal is insufficient to connect patches, and patch size is smaller than the minimum to prevent extinction. If patch size is above the critical size to maintain local populations, the SS > SL pattern is due to the sampling effect of aggregated species distribution combined with the reduction of dispersal ability of species. High‐dispersal rates produce the homogenization of species distribution and a reduction of species richness and this is why habitat fragmentation by habitat loss retains more species when patches are smaller. The model predicts that if a set of species have more dispersal capabilities, the SS effect would be lower than for species with less dispersal. Read the free Plain Language Summary for this article on the Journal blog.

Non-Destructive Estimation of Paper Fiber Using Macro Images: A Comparative Evaluation of Network Architectures and Patch Sizes for Patch-Based Classification

Over the years, research in the field of cultural heritage preservation and document analysis has exponentially grown. In this study, we propose an advanced approach for non-destructive estimation of paper fibers using macro images. Expanding on studies that implemented EfficientNet-B0, we explore the effectiveness of six other deep learning networks, including DenseNet-201, DarkNet-53, Inception-v3, Xception, Inception-ResNet-v2, and NASNet-Large, in conjunction with enlarged patch sizes. We experimentally classified three types of paper fibers, namely, kozo, mitsumata, and gampi. During the experiments, patch sizes of 500, 750, and 1000 pixels were evaluated and their impact on classification accuracy was analyzed. The experiments demonstrated that Inception-ResNet-v2 with 1000-pixel patches achieved the highest patch classification accuracy of 82.7%, whereas Xception with 750-pixel patches exhibited the best macro-image-based fiber estimation performance at 84.9%. Additionally, we assessed the efficacy of the method for images containing text, observing consistent improvements in the case of larger patch sizes. However, limitations exist in background patch availability for text-heavy images. This comprehensive evaluation of network architectures and patch sizes can significantly advance the field of non-destructive paper analysis, offering valuable insights into future developments in historical document examination and conservation science.

Body size and abundance patterns of male orchid bees in a fragmented Neotropical landscape

Abstract Understanding the relationship between functional traits and fragmentation is crucial for promoting biodiversity and ecosystem function. Orchid bees (Euglossini) are important pollinators in Neotropical forests, but how body size may be associated with fragmentation and its influence on community organisation remains unexplored. We systematically sampled orchid bee communities across 15 plots in a fragmented area in the Ecuadorian Amazon basin to evaluate how patch size and isolation are associated with body size and abundance. Our samples comprised large‐bodied bees (Eulaema, Euglossa and the parasitic Exaerete), with small Euglossa notably absent. The relative representation of Eulaema species of varying sizes changed across the fragmented landscape, leading to shifts in overall mean body size consistent with the prediction that larger‐bodied pollinators are better at long‐distance movement. Euglossa, the most abundant genus (67% of the total sample), was found across all patches regardless of body size. Euglossa intersecta and Euglossa orellana, the largest Euglossa in the region, were the most common species across the fragmented landscape. The abundance of male euglossines experienced a general decline with decreasing patch size and increasing isolation. However, at the species level, the abundance of some species was not affected at all. Because body size often mediates dispersal ability, thermoregulation, and resource use, we reinforce the importance of assessing body size patterns to understand changes in diversity, abundance and the composition of tropical pollinator communities in fragmented landscapes.

Optimizing pulmonary chest x-ray classification with stacked feature ensemble and swin transformer integration

This research presents an integrated framework designed to automate the classification of pulmonary chest x-ray images. Leveraging convolutional neural networks (CNNs) with a focus on transformer architectures, the aim is to improve both the accuracy and efficiency of pulmonary chest x-ray image analysis. A central aspect of this approach involves utilizing pre-trained networks such as VGG16, ResNet50, and MobileNetV2 to create a feature ensemble. A notable innovation is the adoption of a stacked ensemble technique, which combines outputs from multiple pre-trained models to generate a comprehensive feature representation. In the feature ensemble approach, each image undergoes individual processing through the three pre-trained networks, and pooled images are extracted just before the flatten layer of each model. Consequently, three pooled images in 2D grayscale format are obtained for each original image. These pooled images serve as samples for creating 3D images resembling RGB images through stacking, intended for classifier input in subsequent analysis stages. By incorporating stacked pooling layers to facilitate feature ensemble, a broader range of features is utilized while effectively managing complexities associated with processing the augmented feature pool. Moreover, the study incorporates the Swin Transformer architecture, known for effectively capturing both local and global features. The Swin Transformer architecture is further optimized using the artificial hummingbird algorithm (AHA). By fine-tuning hyperparameters such as patch size, multi-layer perceptron (MLP) ratio, and channel numbers, the AHA optimization technique aims to maximize classification accuracy. The proposed integrated framework, featuring the AHA-optimized Swin Transformer classifier utilizing stacked features, is evaluated using three diverse chest x-ray datasets-VinDr-CXR, PediCXR, and MIMIC-CXR. The observed accuracies of 98.874%, 98.528%, and 98.958% respectively, underscore the robustness and generalizability of the developed model across various clinical scenarios and imaging conditions.

Long-term patterns of post-fire harvest diverge among ownerships in the Pacific West, U.S.A.

Abstract Post-fire harvest (PFH) is a forest management practice designed to salvage value from burned timber, mitigate safety hazards from dead trees, reduce long-term fuels, and prepare sites for replanting. Despite public controversy and extensive ecological research, little is known about how much PFH occurs on private and public lands in the U.S. Pacific West, or how practices changed with shifting forest policy and increasing area burned over the last three decades. We mapped PFH across 2.2M burned hectares in California, Oregon, and Washington between 1986-2017 and used time series intervention analysis to compare trends in area, rate (% of burned area harvested), and mean patch size between private (0.5M ha) and federal (1.6M ha) forest land and across a gradient of burn severity. Harvest rates varied by ownership (4.9% federal, 18.6% private, 8.0% overall), and practices evolved and diverged over the study period. PFH area and rate declined across all ownerships in the mid-1990s during a period of reduced fire activity. As area burned increased between the early 2000s and late 2010s, PFH area rebounded and surpassed late-1980s levels, while rates remained relatively low. On federal lands, PFH practices shifted in the early-to-mid 1990s towards lower rates (10.3% to 3.8%) and smaller patches (6.0 to 3.3 ha), following policy changes and increased litigation. PFH rates on federal lands decreased at all levels of burn severity, with the largest decreases (6.2% to 1.2%) in forests with low tree mortality (i.e. fire refugia). Conversely, private PFH rates and mean patch sizes more than doubled in forests burned at very low-to-moderate severity. Our results highlight how PFH practices have shifted with policy, socio-economic pressure, and increasing area burned over 31 years in the Pacific West. A similar area of post-fire harvest is now dispersed over larger fires, with practices diverging substantially between ownerships.

Analysis and Visualization of Inter-/Intra-surgeon Variability Using Surgical Simulation for Partial Anomalous Pulmonary Venous Connection Repair.

There may be inter-/intra-surgeon variability in the repair of congenital heart defects. The objective was to analyze and visualize inter-/intra-surgeon variability in the level of patch suture lines and in the shape and size of patches developed through surgical simulation for partial anomalous pulmonary venous connection (PAPVC) repair using a patient-specific 3D-printed heart model. A patient with PAPVC and preoperative computed tomography data were selected, and a patient-specific heart model was obtained. Two different exposures on the model were tested. Two surgeons were enrolled. Both surgeons performed a single-patch repair on four heart models with one exposure and four models with the other. On the potential suture line, 20 points common in each model were allocated. The level of patch suture lines was represented as a deviation from the 20 points. The shape and size of the patches were analyzed and visualized using geometric morphometrics approaches, using the 20 points as landmarks to represent the patches. There was inter-surgeon variability in the level of patch suture lines, and the variability was higher in particular locations. There was inter-surgeon variability in the shape and size of patches, which was not affected by the exposures. The inter-surgeon variability in the shape of patches was higher in particular locations. There was intra-surgeon variability in the shape of patches. Inter-/intra-surgeon variability was analyzed and visualized by the surgical simulation using geometric morphometrics approaches. Further studies are needed to scrutinize whether the variability affects postoperative hemodynamics in actual surgery.

Nonlinear Radiative Response to Patterned Global Warming due to Convection Aggregation and Nonlinear Tropical Dynamics

Abstract Climate responses to global warming exhibit large dependence on the spatial pattern of sea surface temperature (SST) changes. A Green’s function (GF) approach—predicting the global climate response to a complex SST pattern change as the linear superposition of the response to finite-area SST perturbations evaluated in isolation—has been used to systematically evaluate and understand the top of atmosphere (TOA) radiation response to patterned warming. However, in general, the linear GF approach fails for TOA radiation reconstruction under future global warming scenarios in coupled models. Here, we show that the linear superposition of global mean TOA radiation responses to large SST warming perturbations at multiple patches (the GF approach prediction) overestimates the actual response to simultaneous multipatch perturbation. This is because the linear superposition overestimates tropical large-scale convection aggregation strengthening upon localized heating that enhances longwave radiative cooling, which is further explained by the overestimation of circulation response and associated horizontal water vapor transport. The nonadditivity of TOA radiation response is caused by the nonadditivity of convection aggregation, ultimately rooted in nonlinear tropical dynamics. We also demonstrate that the prediction error of the GF approach grows with decreasing patch size (equivalently, increasing patch number). We conclude that using the GF approach to predict future climate change could overestimate longwave radiative cooling and underestimate (effective) climate sensitivity. Numerical experiments may be used to identify specific perturbation patterns where the errors are smaller than the signal. Our research also highlights that an increase in the degree of large-scale convection aggregation has a nonlinear and negative feedback for mean warming.

Jaguar at the Edge: movement patterns in human-altered landscapes

Human-caused habitat loss and fragmentation have significantly impacted the natural environments of large carnivores, altering their movement patterns and increasing risks such as hunting and road collisions. This study aims to understand the movement of jaguars (Panthera onca) through forests, agriculture of varying patch sizes, their distances to these structures, and roads and drainages. By analyzing movement speed, revisits, time spent inside these structures, and the timing of the last visit, data from 54 GPS-tagged jaguars in South America reveal a pronounced tendency to revisit the edges of these landscape variables. Additionally, jaguars showed a stronger affinity for natural areas, spending more time in large forest patches and reducing their speed in natural drainages. Areas with extensive agriculture had fewer revisits, and jaguars moved faster near roads. These results demonstrate the level of tolerance and the dangers this species faces in a landscape with anthropogenic aspects. This comprehensive assessment of movement patterns and landscape use provides valuable insights into how landscape structure influences habitat preference and mobility rates, which is crucial for future jaguar conservation and management strategies.

Plug-and-play adaptive optics microscopy with full-field correction using isoplanatic patch estimation and field segmentation.

We discuss the implementation and performance of a plug-play adaptive optics (AO) module for commercial microscopes comprising indirect wavefront sensing, and a deformable phase plate (DPP) located directly between the objective and the turret. With the DPP at this location, the system closely resembles a pupil-AO scheme, in which effective aberration correction is only possible within the isoplanatic patch. We overcome this limitation by estimating the aberration profiles at multiple field points in parallel and correcting them in sequence to obtain a 2D array of high-quality sub-aperture images. These are then stitched together to form a corrected full-field image. To minimize the measurement time without compromising correction quality, we propose an empirical method to identify the size of the isoplanatic patch, which is both sample and system dependent. Matching the field segment size to that of the isoplanatic patch provides the best compromise between consistent correction quality across the image and measurement time. We demonstrate the performance of the developed system in a commercial microscope using synthetic samples and discuss the performance and limitations of the system.

Patterns of a sexually selected trait are not predictable across urban and non-urban populations of a songbird, the dark-eyed junco

Urbanization can result in novel selective pressures that can cause phenotypic differences amongst urban-tolerant species across urban and non-urban habitats. Here, we compared the size of the white tail patch (“tail white”), a sexual signaling trait, in two urban populations of dark-eyed juncos in comparison to neighboring non-urban populations. Contrary to our expectations, urban phenotypes did not differ from local wildlands in San Diego and Los Angeles counties in similar directions. While the San Diego population showed lower tail white compared to its neighboring wildland population, the Los Angeles population did not. The tail white of the Los Angeles population was not statistically different from that of the San Diego population, suggesting that urban populations may share similar environmental conditions yet face different selective clines due to urbanization. There were, however, differences between wildland populations. Differences in evolutionary histories, environmental conditions, and selective pressures within and outside urban areas may affect how urbanization facilitates population differentiation, even across urban populations of the same species.

Unveiling image source: Instance-level camera device linking via context-aware deep Siamese network

Unveiling the source of an image is one of the most effective ways to validate the originality, authenticity, and reliability in the field of digital forensics. Source camera device identification can identify the specific camera device used to take a photo under investigation. While great progress has been made by the photo-response non-uniformity (PRNU)-based methods over the past decade, the challenge of instance-level source camera device linking, which verifies whether two images in question were captured by the same camera device, remains significant. This challenge is mainly due to the absence of auxiliary images to construct a clean camera fingerprint for each camera, particularly dealing with small image sizes. To overcome this limitation, in this paper, we formulate the task of source device linking as a binary classification problem and propose a simple yet effective framework based on a context-aware deep Siamese network. We take advantage of a Siamese architecture to extract the intrinsic camera device-related noise patterns from a pair of image patches in parallel for comparisons without any auxiliary images. Moreover, a recurrent criss-cross group is utilized to aggregate contextual information in the noise residual maps to alleviate the problem that PRNU noise maps are easily contaminated by the additive noises from image contents. For reliable device linking, we employ a patch-selection strategy on a pair of test images to adaptively choose suitable image patch pairs according to image contents. The final decision of a pair of test images is obtained from the average similarity score of the selected image patch pairs. Compared with existing state-of-the-art methods, our proposed framework can achieve better performance on both the tasks of source camera identification and source device linking without any prior knowledge, i.e., reliable camera fingerprints, regardless of whether the camera devices are “seen” or “unseen” in the training stage. The experimental results on two standard image forensic datasets demonstrate that the proposed method not only shows robustness with respect to different image patch sizes and image quality degenerations, but also has a generalization ability across digital camera and smartphone devices.

Dynamics of a Discrete Population Model

In order to explore the influences of the strong Allee effect on population evolution in two-patch environments, this paper constructs a population model with discrete diffusion and analyzes its dynamics. Using the center manifold theorem, we discuss the codimension-one bifurcation, such as flip bifurcation, of the model. The qualitative structures and stability of the model at the degenerate fixed point are obtained from the approximation of flow. Based on the theoretical analysis results, we investigate the influences of the strong Allee effect and diffusion on population evolution. The qualitative structures of degenerate fixed point and numerical simulation show that the strong Allee effect will make the population tend to a stable size when the two initial population patch sizes are equal and exceed a threshold. But the stable size of population does not exceed the carrying capacity, depending on the constant Allee effect and carrying capacity, which differ from the conclusion of logistic model that population size tends to the carrying capacity. However, if the two initial population patches are equal in size and below this threshold, or if the sizes of the initial population patches are unequal, the population size will decrease. The fixed point becomes degenerate with respect to the strong Allee effect causing the population tend to a stable size below carrying capacity, which exacerbates the complexity of population evolution.

Habitat Fragmentation Affects Richness—A View Through a Metacommunity Lens

Habitat fragmentation is often assumed to negatively impact species diversity because smaller, more isolated populations on smaller habitat patches are at a higher extinction risk. However, some empirical and theoretical studies suggest that landscapes with numerous small habitat patches may support higher species richness, although the circumstances remain elusive. We used an agent-based metacommunity model to investigate this and simulate landscapes of the same total area but diverse patch sizes. Our model, as generic and unbiased by specific assumptions as possible, aimed to explore which circumstances may be more conducive to supporting higher biodiversity. To this end, most parameters and behaviors were random. The model included generalized species traits, dispersal, and interactions to explore species richness dynamics in fragmented landscapes of distinct patch sizes. Our results show that landscapes with many small patches maintain higher species richness than those with fewer large patches. Moreover, the relationship between patch connectivity and species richness is more pronounced in landscapes with smaller patches. High connectivity in these landscapes may support species diversity by preventing local extinctions and facilitating recolonization. In contrast, connectivity is less significant in large-patch landscapes, where generalist species dominate. The findings highlight the complex interplay between patch size quality, connectivity, species traits, and diverse interactions among species in determining species richness. We suggest the patterns produced by the model represent null predictions and may be useful as a reference for a diversity of more specialized questions and predictions. These insights may also have specific implications for conservation strategies, suggesting that maintaining a mosaic of small, well-connected patches could enhance biodiversity in fragmented landscapes.

Precise grading of non-muscle invasive bladder cancer with multi-scale pyramidal CNN

The grading of non-muscle invasive bladder cancer (NMIBC) continues to face challenges due to subjective interpretations, which affect the assessment of its severity. To address this challenge, we are developing an innovative artificial intelligence (AI) system aimed at objectively grading NMIBC. This system uses a novel convolutional neural network (CNN) architecture called the multi-scale pyramidal pretrained CNN to analyze both local and global pathology markers extracted from digital pathology images. The proposed CNN structure takes as input three levels of patches, ranging from small patches (e.g., 128×128\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$128 \ imes 128$$\\end{document}) to the largest size patches (512×512\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$512 \ imes 512$$\\end{document}). These levels are then fused by random forest (RF) to estimate the severity grade of NMIBC. The optimal patch sizes and other model hyperparameters are determined using a grid search algorithm. For each patch size, the proposed system has been trained on 32K patches (comprising 16K low-grade and 16K high-grade samples) and subsequently tested on 8K patches (consisting of 4K low-grade and 4K high-grade samples), all annotated by two pathologists. Incorporating light and efficient processing, defining new benchmarks in the application of AI to histopathology, the ShuffleNet-based AI system achieved notable metrics on the testing data, including 94.25% ± 0.70% accuracy, 94.47% ± 0.93% sensitivity, 94.03% ± 0.95% specificity, and a 94.29% ± 0.70% F1-score. These results highlight its superior performance over traditional models like ResNet-18. The proposed system’s robustness in accurately grading pathology demonstrates its potential as an advanced AI tool for diagnosing human diseases in the domain of digital pathology.