Explore the vast range of titles available.

We propose a technique for producing ‘visual explanations’ for decisions from a large class of Convolutional Neural Network (CNN)-based models, making them more transparent and explainable. Our approach—Gradient-weighted Class Activation Mapping (Grad-CAM), uses the gradients of any target concept (say ‘dog’ in a classification network or a sequence of words in captioning network) flowing into the final convolutional layer to produce a coarse localization map highlighting the important regions in the image for predicting the concept. Unlike previous approaches, Grad-CAM is applicable to a wide variety of CNN model-families: (1) CNNs with fully-connected layers (e.g.VGG), (2) CNNs used for structured outputs (e.g.captioning), (3) CNNs used in tasks with multi-modal inputs (e.g.visual question answering) or reinforcement learning, all without architectural changes or re-training. We combine Grad-CAM with existing fine-grained visualizations to create a high-resolution class-discriminative visualization, Guided Grad-CAM, and apply it to image classification, image captioning, and visual question answering (VQA) models, including ResNet-based architectures. In the context of image classification models, our visualizations (a) lend insights into failure modes of these models (showing that seemingly unreasonable predictions have reasonable explanations), (b) outperform previous methods on the ILSVRC-15 weakly-supervised localization task, (c) are robust to adversarial perturbations, (d) are more faithful to the underlying model, and (e) help achieve model generalization by identifying dataset bias. For image captioning and VQA, our visualizations show that even non-attention based models learn to localize discriminative regions of input image. We devise a way to identify important neurons through Grad-CAM and combine it with neuron names (Bau et al. in Computer vision and pattern recognition, 2017) to provide textual explanations for model decisions. Finally, we design and conduct human studies to measure if Grad-CAM explanations help users establish appropriate trust in predictions from deep networks and show that Grad-CAM helps untrained users successfully discern a ‘stronger’ deep network from a ‘weaker’ one even when both make identical predictions. Our code is available at https://github.com/ramprs/grad-cam/, along with a demo on CloudCV (Agrawal et al., in: Mobile cloud visual media computing, pp 265–290. Springer, 2015) (http://gradcam.cloudcv.org) and a video at http://youtu.be/COjUB9Izk6E.

Despite extensive studies into severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), the effect of maternal infection on the neonate is unclear. To investigate this, we characterized the immunology of neonates born to mothers with confirmed SARS-CoV-2 infection during pregnancy. Here we show that maternal SARS-CoV-2 infection affects the neonatal immune system. Despite similar proportions of B cells, CD4 + T cells and CD8 + T cells, increased percentages of natural killer cells, Vδ2 + γδ T cells and regulatory T cells were detected in neonates born to mothers with recent or ongoing infection compared with those born to recovered or uninfected mothers. Increased plasma cytokine levels were also evident in neonates and mothers within the recent or ongoing infection group. Cytokine functionality was enhanced in neonates born to SARS-CoV-2-exposed mothers, compared to those born to uninfected mothers. In most neonates, this immune imprinting was nonspecific, suggesting vertical transmission of SARS-CoV-2 is limited, a finding supported by a lack of SARS-CoV-2-specific IgM in neonates despite maternal IgG transfer. Many babies have now been born to mothers who were exposed to SARS-CoV-2 during their pregnancy. Here the authors look at the effect of this exposure on the immunology of human neonates, showing immune changes and increased neonatal cytokine responses despite limited evidence of vertical transmission.

Accurate localization and segmentation of the optic disc (OD) are considered crucial for the early detection of ophthalmic diseases such as glaucoma and diabetic retinopathy. Challenges such as image quality variability, high background noise, and insufficient edge information are often encountered by existing methods. To address these issues, an adaptive framework is proposed in which Fast Circlet Transformation (FCT) is combined with entropy-based features derived from retinal blood vessels for robust OD localization. Minkowski weighted K-means clustering is utilized to dynamically assess feature importance, thereby enhancing resilience to dataset variations. Following localization, partial differential equation-based image inpainting is employed for blood vessel removal, and OD segmentation is refined using the Chan-Vese active contour model. The method’s localization efficacy is demonstrated through extensive evaluations across multiple public datasets (DRISHTI-GS, DRIONS-DB, IDRID, and ORIGA), and segmentation performance metrics, including Dice coefficients of 0.94–0.95 and Jaccard indices of 0.9, are achieved on the ORIGA and DRISHTI-GS datasets. Through these results, the robustness and generalizability of the proposed method for clinical applications in retinal image analysis are highlighted.

This paper investigates the effects of process parameters on the joint strength and process robustness when multi-layered joints of dissimilar metals are produced by ultrasonic metal welding (UMW). Three layers of 0.3-mm aluminium sheet are welded with a single 1.0-mm copper sheet which is representative of electric vehicle battery interconnects. A process robustness study in which welding pressure, amplitude of vibration and welding time are varied to produce satisfactory welds is reported. The weld quality is evaluated by performing lap shear and T-peel tests where maximum loads are considered as the quality indicator. Response surfaces are developed to identify the relationship and sensitivity between the input process parameters and output quality indicators. A feasible weldability zone is defined for the first time by identifying the under-weld, good-weld and over-weld conditions based on load-displacement curves and corresponding failure modes. Relying on the weldability zone and response surfaces, multi-objective optimisation is performed to obtain maximum lap shear and T-peel strength which resulted in Pareto frontier or trade-off curve between both objectives. An optimal joint is selected from the Pareto front which is verified and validated by performing confirmation experiments, and further, used for T-peel strength analysis of different interfaces of the multi-layered joint. To conclude, this paper determines both the optimal weld parameters and the robust operating range.

Mobile Crowdsensing (MCS) has shown the greatest potential that allows smart devices to collect and share different sensing data. Mobile users (or participants) send the desired sensing data to the service providers and collect rewards. However, the reward needs to be given such as, it does not increase platform costs. On the other hand, the unsatisfactory reward may reduce the interest of the participant which may degrade the quality data. Therefore, increasing sensing data quality with a constrained budget is a crucial challenge. There has been extensive research on the reward mechanism for MCS, but, most of the work is on the basic assumption that participant will complete their assigned task positively. In this paper, we propose an efficient user selection mechanism for Mobile Crowdsensing System (MCS) by considering the Probability of Success (PoS) of users (i.e. participant may fail to complete the assigned task with some probability). For the selection of an efficient user, the proposed mechanism also accounts the parameters like data quality and platform cost. We also propose a reward calculation model for the selected users. Minimizing the platform cost with a constrained budget is an NP-hard problem. To provide a sub-optimal solution to this problem Chaotic Krill-Herd optimization algorithm is used. The extensive simulation results reveal that the proposed method outperforms the existing work by considerable margins.

Air quality forecasting is an important application area of the time series forecasting problem. The successful prediction of the air quality of a place well in advance can able to help administrators to take the necessary steps to control air pollution. The administrator can also warn the citizens about the adverse effect of air pollution in advance. In this study, an air quality forecasting method is proposed to successfully forecast the air quality of a place. Here the type-2 fuzzy time series (FTS) forecasting method is applied to predict air quality. The performance of any FTS heavily depends on the selection of its hyperparameters. In this letter, a fuzzy time series optimization (FTSBO) algorithm is proposed to optimize all the hyperparameters of the FTS forecasting method. The proposed FTSBO algorithm originated from the butterfly optimization technique. In this work, the performance of the proposed forecasting method is also compared to the well-known forecasting methods. The simulation results established that the proposed forecasting method produces satisfactory performance, and its performance is better in comparison to other well-known forecasting methods.

In order to migrate over large distances, cells within tissues and organisms rely on sensing local gradient cues which are irregular, conflicting, and changing over time and space. The mechanism how they generate persistent directional migration when signals are disrupted, while still remaining adaptive to signal’s localization changes remain unknown. Here, we find that single cells utilize a molecular mechanism akin to a working memory to satisfy these two opposing demands. We derive theoretically that this is characteristic for receptor networks maintained away from steady states. Time-resolved live-cell imaging of Epidermal growth factor receptor (EGFR) phosphorylation dynamics shows that cells transiently memorize position of encountered signals via slow-escaping remnant of the polarized signaling state, a dynamical ‘ghost’, driving memory-guided persistent directional migration. The metastability of this state further enables migrational adaptation when encountering new signals. We thus identify basic mechanism of real-time computations underlying cellular navigation in changing chemoattractant fields. If we are injured, or fighting an infection, cells in our body migrate over large distances to the site of the wound or infection to act against any invading microbes or repair the damage. Cells navigate to the damaged site by sensing local chemical cues, which are irregular, conflicting and change over time and space. This implies that cells can choose which direction to travel, and stick to it even if the signals around them are disrupted, while still retaining the ability to alter their direction if the location of the signal changes. However, how cells are able to effectively navigate their way through this field of complex chemical cues is poorly understood. To help resolve this mystery, Nandan, Das et al. studied the epidermal growth factor receptor (EGFR) signaling network which controls how some cells in the body change shape and migrate. The network is activated by specific chemical cues, or ligands, binding to EGFR proteins on the cell surface. The receptors then join together to form pairs, and several tags known as phosphate groups are added to each molecule. This process (known as phosphorylation) switches the receptor pair to an active state, allowing EGFR to relay signals to other proteins in the cell and promote the activity of receptors not bound to a ligand. The phosphorylation state of EGFRs is then modulated over time and across the cell by a network of enzymes called phosphatases which can remove the phosphate groups and switch off the receptor. To study EGFR phosphorylation dynamics in human cells, Nandan, Das et al. imaged individual cells over time using a microscope. This data was then combined with a mathematical model describing the EGFR signaling network and how cells change their shape over time. The experiment revealed that the phosphate groups attached to EGFR are not removed immediately when the chemical cue is gone. Instead, the active state is transiently maintained before complete inactivation. This had the effect of encoding a short-lived memory in the signaling network that allowed the cells to continue to migrate in a certain direction even when chemical cues were disrupted . This memory state is dynamic, enabling cells to adapt direction when the cue changes location. The findings of Nandan, Das et al. reveal the underlying mechanism for how cells decipher complex chemical cues to migrate to where they are needed most. The next steps to follow on from this work will be to understand if other receptors involved in migration work in a similar way.

Retinal diseases such as Diabetic Retinopathy (DR), Hypertensive Retinopathy (HR), different types of Occlusions, etc., are associated with the deformity observed in the Retinal Vessel Structure (RVS). This paper proposes an automatic unsupervised vessel segmentation technique to separate the RVS with insignificant change in curvature of the vessel and eliminate the noises from the vessel structure and the background. The method involves three phases: preprocessing, where the fundus image is enhanced based on local information, and the noises are separated from the vessels. The second phase introduces a unique Bel–Hat transformation, which simultaneously uses two different groups of Structural Elements: the Neighbor Adaptive Line Structuring Element (NALSE) and the 2D Gaussian Structuring Element (2DGSE). These combined groups of Structural Elements can separate the vessel structure from the background by changing the size and orientation of the Structural Elements. Lastly, a novel robust statistical threshold is used, based on the statistical distribution of the area of the isolated objects, to segment the accurate noise-free Retinal Vessel Structure (RVS). This proposed method is more accurate than the recently proposed unsupervised and supervised methods.

Proton pump inhibitors (PPIs) are among the most prescribed drugs worldwide owing to their proven efficacy in symptom control and mucosal healing for acid-related disorders including gastroesophageal reflux disease (GORD), peptic ulcer disease, Helicobacter pylori eradication, functional dyspepsia, and gastroprotection in high-risk patients. However, long-term use beyond approved indications is increasingly common and has raised safety concerns. Observational studies link chronic PPI use to a myriad of adverse outcomes such as enteric infections (e.g., Clostridioides difficile), nutrient deficiencies (magnesium, vitamin B12), osteoporotic fractures, chronic kidney disease, dementia, and gastric and colorectal cancer. While causality is not always established, these associations warrant cautious risk–benefit assessment in patients receiving prolonged therapy. Current guidelines advocate periodic review of ongoing PPI use and emphasise deprescribing where appropriate. Strategies include dose reduction, on-demand or intermittent use, and switching to H2-receptor antagonists, particularly in patients with non-erosive reflux disease or functional dyspepsia. Tools from the National Institute for Health and Clinical Excellence, American College of Gastroenterology, and the Canadian Deprescribing Network assist clinicians in identifying candidates for tapering or discontinuation. This narrative review focuses on the concept of “PPI stewardship” by providing an evidence-based overview of PPI indications, risks, and deprescribing strategies to promote appropriate, safer, and patient-centred use of acid-suppressive therapy.