Acute Coronary Syndromes (ACS), including ST- and non-ST-segment elevation myocardial infarction (STEMI, NSTEMI), remain a leading cause of global mortality. Traditional Cardiovascular Risk Scores (CVRS) provide important insights but mainly rely on clinical data, often neglecting environmental factors (e.g.air pollution, climate) that significantly influence cardiovascular health. Integrating complex time-series environmental and clinical datasets also presents substantial challenges. We propose TabulaTime, a multimodal deep learning framework integrating clinical risk factors with environmental data to enhance ACS risk prediction. TabulaTime delivers three innovations: multimodal integration of time-series environmental and clinical data; PatchRWKV for extracting complex temporal patterns with linear computational complexity; and enhanced interpretability through attention mechanisms. TabulaTime improves prediction accuracy by 20.5% over CatBoost, with environmental data contributing a 10.1% gain. PatchRWKV outperforms state-of-the-art models (MLP-, CNN-, RNN- and Transformer-based models). Feature analysis highlights key clinical and environmental predictors. This approach advances personalised prevention and strengthens public health against cardiovascular risks. • A novel multimodal deep learning framework, TabulaTime, integrates clinical and environmental time-series data to improve Acute Coronary Syndrome (ACS) risk prediction. • We introduce PatchRWKV, an efficient time-series feature extractor with linear complexity that outperforms state-of-the-art models in capturing temporal patterns. • The integration of environmental data improves ACS prediction accuracy by 10.1%, with feature analysis identifying key clinical and pollution-related predictors.

Accurate diagnosis of diabetic retinopathy increasingly depends on the complementary strengths of multi-modal imaging, particularly colour fundus photography (CFP) and optical coherence tomography (OCT). However, most existing selfsupervised learning (SSL) methods focus on single modalities and rely on either paired data or extensive manual annotations. These requirements limit their scalability and practical use in clinical settings. To address these challenges, we propose Align4Eye, a novel SSL framework that learns unified representations from unlabelled and unpaired CFP and OCT images. Instead of relying on direct image correspondences, strong supervision, or instancelevel contrastive learning, Align4Eye introduces a lightweight, symmetric alignment mechanism integrated with masked image modelling. Specifically, the model computes per-modality feature representations and aligns them to a shared set of semantic prototypes within a common embedding space. This promotes semantic consistency across modalities while preserving modalityspecific details, enabling the system to capture disease-relevant patterns without requiring labels or cross-modal pairing. We evaluate Align4Eye on three widely used retinal datasets and show that it outperforms existing SSL baselines on APTOS and OCTiD, while achieving comparable performance on Messidor-2. The method improves both accuracy and AUC by up to 8 % over ImageNet pretraining, and achieves$2 - 1 0 \%$gains over strong SSL baselines such as MAE, BYOL, and Uni4Eye in APTOS and OCTiD evaluations. Additional ablation studies further demonstrate the effectiveness of the proposed alignment strategy in enabling robust and generalisable multi-modal representation learning.

Depth estimation is a critical task in computer vision, with applications in autonomous navigation, robotics, and augmented reality. Event cameras, which encode temporal changes in light intensity as asynchronous binary spikes, offer unique advantages such as low latency, high dynamic range, and energy efficiency. However, their unconventional spiking output and the scarcity of labelled datasets pose significant challenges to traditional image-based depth estimation methods. To address these challenges, we propose a novel energy-efficient Spike-Driven Transformer Network (SDT) for depth estimation, leveraging the unique properties of spiking data. The proposed SDT introduces three key innovations: (1) a purely spike-driven transformer architecture that incorporates spike-based attention and residual mechanisms, enabling precise depth estimation with minimal energy consumption; (2) a fusion depth estimation head that combines multi-stage features for fine-grained depth prediction while ensuring computational efficiency; and (3) a cross-modality knowledge distillation framework that utilises a pre-trained vision foundation model (DINOv2) to enhance the training of the spiking network despite limited data availability. Experimental evaluations on synthetic and real-world event datasets demonstrate the superiority of our approach, with substantial improvements in Absolute Relative Error (49 % reduction) and Square Relative Error (39.77 % reduction) compared to existing models. The SDT also achieves a 70.2 % reduction in energy consumption (12.43 mJ vs. 41.77 mJ per inference) and reduces model parameters by 42.4 % (20.55 M vs. 35.68 M), making it highly suitable for resource-constrained environments. This work represents the first exploration of transformer-based spiking neural networks for depth estimation, providing a significant step forward in energy-efficient neuromorphic computing for real-world vision applications.

Deep learning (DL) has significantly advanced super-resolution (SR), a technique that enhances low-quality images by reconstructing fine details. However, most DL-based SR methods struggle at high magnification levels (e.g., ×4 or higher) due to dramatically increased ill-posedness. To overcome this, we define high-magnification SR as an ultra-resolution (UR) problem and introduce WaveDiffUR, a novel wavelet-domain diffusion model designed for extreme-scale image reconstruction. WaveDiffUR decomposes the UR process into sequential steps, first restoring low-frequency wavelet details for global consistency and then refining high-frequency components for sharper textures. By integrating pre-trained SR models as modular components, it reduces ill-posedness and ensures adaptability across different applications. Unlike existing SR approaches, which struggle with fixed boundary conditions at extreme magnifications, WaveDiffUR incorporates the cross-scale pyramid (CSP) constraint, an adaptive framework that dynamically refines low- and high-frequency wavelet details to maintain consistency and high fidelity. Extensive experiments demonstrate that WaveDiffUR with CSP notably enhances spatial accuracy and consistently generates high-frequency details with remarkable fidelity during the SR process. Evaluations are conducted across two benchmark evaluation datasets and four additional independent datasets. The empirical results reveal that, as magnification scales from ×8 to ×128, WaveDiffUR achieves an average degradation rate in PSNR, NIQE, and SRE of only 19.1%—the best performance among all benchmarked models—while consistently delivering sharper images characterized by superior spatial fidelity. By enabling scalable, high-fidelity ultra-resolution, WaveDiffUR opens new possibilities for remote sensing applications, including environmental monitoring, urban planning, disaster response, and precision agriculture.

Potatoes, a crucial staple crop, face significant threats from late blight, which poses serious risks to food security. Despite extensive research using ground and unmanned aerial vehicle (UAV) hyperspectral data for crop disease monitoring, satellite-scale identification of diseases, such as potato late blight (PLB) remains limited. This study employs a multiscale analysis approach, integrating high-resolution Sentinel-2 multispectral satellite data with UAV and ground spectral data, to monitor and identify PLB. A key finding of this study is the general similarity in spectral patterns across different scales, with consistent valley values in bands of blue and red and peak values in bands of near infrared (NIR) and narrow NIR, accompanied by a consistent decrease in reflectance correlating with increasing disease severity. Furthermore, the study highlights scale-dependent spectral variations, with changes in bands of Vegetation Red Edge2, Vegetation Red Edge3, NIR, and narrow NIR being more pronounced at the ground scale compared to UAV and satellite scales. Based on the developed red edge index and disease stress index with a suite of machine learning algorithms, we proposed an XGBoost-based model integrating spectral indices for PLB monitoring (PLB-SI-XGBoost). Notably, the proposed model demonstrated the highest average evaluation score of 0.88 and the lowest root-mean-square error (RMSE) of 13.50 during ground-scale validation, outperforming other algorithms. At the UAV scale, the proposed model achieved a robust R-squared value of 0.74 and an RMSE of 18.27. Moreover, the application of Sentinel-2 data for disease detection at the satellite scale yielded an accuracy of 70% in the model. The results of the study emphasize the importance of scale in disease monitoring models and illuminate the potential for satellite-scale surveillance of PLB. The exceptional performance of the PLB-SI-XGBoost model in detecting PLB suggests its utility in enhancing agricultural decision-making with more accurate and reliable data support.

Deep Neural Network (DNN) models are usually trained sequentially from one layer to another, which causes forward, backward and update locking problems, leading to poor performance in terms of training time. The existing parallel strategies to mitigate these problems provide suboptimal runtime performance. In this work, we have proposed a novel layer-wise partitioning and merging, forward and backward pass parallel framework to provide better training performance. The novelty of the proposed work consists of 1) a layer-wise partition and merging model which can minimise communication overhead between devices without the memory cost of existing strategies during the training process; 2) a forward pass and backward pass parallelisation to address the update locking problem and minimise the total training cost. The experimental evaluation on real use cases shows that the proposed method outperforms the state-of-the-art approaches in terms of training speed; and achieves almost linear speedup without compromising the accuracy performance of the non-parallel approach.

Accurate and rapid detection of COVID-19 pneumonia is crucial for optimal patient treatment. Chest X-Ray (CXR) is the first-line imaging technique for COVID-19 pneumonia diagnosis as it is fast, cheap and easily accessible. Currently, many deep learning (DL) models have been proposed to detect COVID-19 pneumonia from CXR images. Unfortunately, these deep classifiers lack the transparency in interpreting findings, which may limit their applications in clinical practice. The existing explanation methods produce either too noisy or imprecise results, and hence are unsuitable for diagnostic purposes. In this work, we propose a novel explainable CXR deep neural Network (CXR-Net) for accurate COVID-19 pneumonia detection with an enhanced pixel-level visual explanation using CXR images. An Encoder-Decoder-Encoder architecture is proposed, in which an extra encoder is added after the encoder-decoder structure to ensure the model can be trained on category samples. The method has been evaluated on real world CXR datasets from both public and private sources, including healthy, bacterial pneumonia, viral pneumonia and COVID-19 pneumonia cases. The results demonstrate that the proposed method can achieve a satisfactory accuracy and provide fine-resolution activation maps for visual explanation in the lung disease detection. Compared to current state-of-the-art visual explanation methods, the proposed method can provide more detailed, high-resolution, visual explanation for the classification results. It can be deployed in various computing environments, including cloud, CPU and GPU environments. It has a great potential to be used in clinical practice for COVID-19 pneumonia diagnosis.

Spectral–spatial-based deep learning models have recently proven to be effective in hyper-spectral image (HSI) classification for various earth monitoring applications such as land cover classification and agricultural monitoring. However, due to the nature of “black-box” model representation, how to explain and interpret the learning process and the model decision, especially for vegetation classification, remains an open challenge. This study proposes a novel interpretable deep learning model—a biologically interpretable two-stage deep neural network (BIT-DNN), by incorporating the prior-knowledge (i.e., biophysical and biochemical attributes and their hierarchical structures of target entities)-based spectral–spatial feature transformation into the proposed framework, capable of achieving both high accuracy and interpretability on HSI-based classification tasks. The proposed model introduces a two-stage feature learning process: in the first stage, an enhanced interpretable feature block extracts the low-level spectral features associated with the biophysical and biochemical attributes of target entities; and in the second stage, an interpretable capsule block extracts and encapsulates the high-level joint spectral–spatial features representing the hierarchical structure of biophysical and biochemical attributes of these target entities, which provides the model an improved performance on classification and intrinsic interpretability with reduced computational complexity. We have tested and evaluated the model using four real HSI data sets for four separate tasks (i.e., plant species classification, land cover classification, urban scene recognition, and crop disease recognition tasks). The proposed model has been compared with five state-of-the-art deep learning models. The results demonstrate that the proposed model has competitive advantages in terms of both classification accuracy and model interpretability, especially for vegetation classification.

Realistic hyperspectral image (HSI) super-resolution (SR) techniques aim to generate a high-resolution (HR) HSI with higher spectral and spatial fidelity from its low-resolution (LR) counterpart. The generative adversarial network (GAN) has proven to be an effective deep learning framework for image super-resolution. However, the optimisation process of existing GAN-based models frequently suffers from the problem of mode collapse, leading to the limited capacity of spectral-spatial invariant reconstruction. This may cause the spectral-spatial distortion on the generated HSI, especially with a large upscaling factor. To alleviate the problem of mode collapse, this work has proposed a novel GAN model coupled with a latent encoder (LE-GAN), which can map the generated spectral-spatial features from the image space to the latent space and produce a coupling component to regularise the generated samples. Essentially, we treat an HSI as a high-dimensional manifold embedded in a latent space. Thus, the optimisation of GAN models is converted to the problem of learning the distributions of high-resolution HSI samples in the latent space, making the distributions of the generated super-resolution HSIs closer to those of their original high-resolution counterparts. We have conducted experimental evaluations on the model performance of super-resolution and its capability in alleviating mode collapse. The proposed approach has been tested and validated based on two real HSI datasets with different sensors (i.e. AVIRIS and UHD-185) for various upscaling factors (i.e. ×2, ×4, ×8) and added noise levels (i.e. ∞ db, 40 db, 80 db), and compared with the state-of-the-art super-resolution models (i.e. HyCoNet, LTTR, BAGAN, SR- GAN, WGAN). Experimental results show that the proposed model outperforms the competitors on the super-resolution quality, robustness, and alleviation of mode collapse. The proposed approach is able to capture spectral and spatial details and generate more faithful samples than its competitors. It has also been found that the proposed model is more robust to noise and less sensitive to the upscaling factor and has been proven to be effective in improving the convergence of the generator and the spectral-spatial fidelity of the super-resolution HSIs.

Computer-aided early diagnosis of Alzheimer's disease (AD) and its prodromal form mild cognitive impairment (MCI) based on structure Magnetic Resonance Imaging (sMRI) has provided a cost-effective and objective way for early prevention and treatment of disease progression, leading to improved patient care. In this work, we have proposed a novel computer-aided approach for early diagnosis of AD by introducing an explainable 3D Residual Attention Deep Neural Network (3D ResAttNet) for end-to-end learning from sMRI scans. Different from the existing approaches, the novelty of our approach is three-fold: 1) A Residual Self-Attention Deep Neural Network has been proposed to capture local, global and spatial information of MR images to improve diagnostic performance; 2) An explainable method using Gradient-based Localization Class Activation mapping (Grad-CAM) has been introduced to improve the interpretability of the proposed method; 3) This work has provided a full end-to-end learning solution for automated disease diagnosis. Our proposed 3D ResAttNet method has been evaluated on a large cohort of subjects from real datasets for two changeling classification tasks (i.e. Alzheimer's disease (AD) vs. Normal cohort (NC) and progressive MCI (pMCI) vs. stable MCI (sMCI)). The experimental results show that the proposed approach has a competitive advantage over the state-of-the-art models in terms of accuracy performance and generalizability. The explainable mechanism in our approach is able to identify and highlight the contribution of the important brain parts (e.g., hippocampus, lateral ventricle and most parts of the cortex) for transparent decisions.

Structural magnetic resonance imaging (sMRI) is widely used for the brain neurological disease diagnosis, which could reflect the variations of brain. However, due to the local brain atrophy, only a few regions in sMRI scans have obvious structural changes, which are highly correlative with pathological features. Hence, the key challenge of sMRI-based brain disease diagnosis is to enhance the identification of discriminative features. To address this issue, we propose a dual attention multi-instance deep learning network (DA-MIDL) for the early diagnosis of Alzheimer's disease (AD) and its prodromal stage mild cognitive impairment (MCI). Specifically, DA-MIDL consists of three primary components: 1) the Patch-Nets with spatial attention blocks for extracting discriminative features within each sMRI patch whilst enhancing the features of abnormally changed micro-structures in the cerebrum, 2) an attention multi-instance learning (MIL) pooling operation for balancing the relative contribution of each patch and yield a global different weighted representation for the whole brain structure, and 3) an attention-aware global classifier for further learning the integral features and making the AD-related classification decisions. Our proposed DA-MIDL model is evaluated on the baseline sMRI scans of 1689 subjects from two independent datasets (i.e., ADNI and AIBL). The experimental results show that our DA-MIDL model can identify discriminative pathological locations and achieve better classification performance in terms of accuracy and generalizability, compared with several state-of-the-art methods.

Data-driven intelligence can play a pivotal role in enhancing the effectiveness and efficiency of police service provision. Despite of police organizations being a rich source of qualitative data (present in less formally structured formats, such as the text logs), little work has been done in automating steps to allow this data to feed into intelligence-led policing tasks, such as demand analysis/prediction. This paper examines the use of police incident logs to better estimate the demand of officers across all incidents, with particular respect to the cases where mental-ill health played a primary part. Persons suffering from mental-ill health are significantly more likely to come into contact with the police, but statistics relating to how much actual police time is spent dealing with this type of incident are highly variable and often subjective. We present a novel deep learning based text mining approach, which allows accurate extraction of mental-ill health related incidents from police incident logs. The data gained from these automated analyses can enable both strategic and operational planning within police forces, allowing policy makers to develop long term strategies to tackle this issue, and to better plan for day-today demand on services. The proposed model has demonstrated the cross-validated classification accuracy of 89.5% on the real dataset.

Summary The past few years have witnessed a rapid deployment of computing infrastructures in the cloud in support of data intensive applications. The effort of the existing works is mainly focused on data reusing mechanisms without considering data processing routes, which can significantly affect the computation costs when exchanging data among the computing node in the cloud. This paper presents a genetic algorithm enhanced Automatic Data Flow Management Solution (ADFMS) that facilitates automatic routing function and a self‐adjustable intermediate data management mechanism to achieve an efficient data processing structure of cloud computing. Experimental results show that ADFMS optimizes costs in managing intermediate data in the cloud.

Scanning laser ophthalmoscopes (SLOs) can be used for early detection of retinal diseases. With the advent of latest screening technology, the advantage of using SLO is its wide field of view, which can image a large part of the retina for better diagnosis of the retinal diseases. On the other hand, during the imaging process, artefacts such as eyelashes and eyelids are also imaged along with the retinal area. This brings a big challenge on how to exclude these artefacts. In this paper, we propose a novel approach to automatically extract out true retinal area from an SLO image based on image processing and machine learning approaches. To reduce the complexity of image processing tasks and provide a convenient primitive image pattern, we have grouped pixels into different regions based on the regional size and compactness, called superpixels. The framework then calculates image based features reflecting textural and structural information and classifies between retinal area and artefacts. The experimental evaluation results have shown good performance with an overall accuracy of 92%.

This paper proposes a new algorithm-FC-WPath, a fuzzy rule-based classification of weighted path traversals using fuzzy quantifiers for web usage mining. Web usage mining usually analyses frequent path traversals or frequent subgraphs where each path has the same level of importance. However, the current frequent pattern mining based methods could not distinguish the level of importance for different paths. Further, there is little work done in relation to classification of path traversals based on fuzzy classification inferences and fuzzy quantification, which often provides good readability and interpretation of complex patterns. In this work, we attach numeric weights to each path traversed according to some level of importance, therefore introducing quantitative and fuzzy values. The derived fuzzy if-then classification rules from weighted paths can then be described both by the linguistic fuzzy rules and linguistic quantifiers like "all", "some" etc. As a result, we propose a fuzzy subset-hood model with fuzzy quantifiers for describing the usual fuzzy if-then rules applied to web usage mining. The experimental result shows that the proposed FC-WPath algorithm has good classification accuracy, readability and runtime.

Low back pain (LBP) affects most people at some time in their life and psychological factors are often viewed as obstacles to recovery. LBP is often accompanied by hyperactivity of superficial Para spinal muscles and it has been suggested that psychological factors may affect the condition via increased spinal loading resulting from altered Para spinal muscle activity. Several measurements are taken, including physical factors (muscle activity, pain intensity, disability) and psychosocial factors (anxiety, depression, fear of movement etc) using several numerical scales and questionnaires. The aim of this work is to obtain relationships between these measurements. Most data recorded for LBP is numerical and range bound (intervalised or scaled), therefore presenting inherent fuzziness. To find relationships in the data, we have used a fuzzy association rule mining approach to identify correlations. Further, the use of fuzzy terms (linguistic terms) in the generated fuzzy rules helps to interpret the clinical outcome (readability). To show the applicability of this method, we have conducted experiments on a real LBP clinical dataset which indicate both valuable associations and understandable and interpretable rules. The results have been validated by an exercise and sports scientist.

MOTIVATION: Deciphering the regulatory and developmental mechanisms for multicellular organisms requires detailed knowledge of gene interactions and gene expressions. The availability of large datasets with both spatial and ontological annotation of the spatio-temporal patterns of gene expression in mouse embryo provides a powerful resource to discover the biological function of embryo organization. Ontological annotation of gene expressions consists of labelling images with terms from the anatomy ontology for mouse development. If the spatial genes of an anatomical component are expressed in an image, the image is then tagged with a term of that anatomical component. The current annotation is done manually by domain experts, which is both time consuming and costly. In addition, the level of detail is variable, and inevitably errors arise from the tedious nature of the task. In this article, we present a new method to automatically identify and annotate gene expression patterns in the mouse embryo with anatomical terms. RESULTS: The method takes images from in situ hybridization studies and the ontology for the developing mouse embryo, it then combines machine learning and image processing techniques to produce classifiers that automatically identify and annotate gene expression patterns in these images. We evaluate our method on image data from the EURExpress study, where we use it to automatically classify nine anatomical terms: humerus, handplate, fibula, tibia, femur, ribs, petrous part, scapula and head mesenchyme. The accuracy of our method lies between 70% and 80% with few exceptions. We show that other known methods have lower classification performance than ours. We have investigated the images misclassified by our method and found several cases where the original annotation was not correct. This shows our method is robust against this kind of noise. AVAILABILITY: The annotation result and the experimental dataset in the article can be freely accessed at http://www2.docm.mmu.ac.uk/STAFF/L.Han/geneannotation/. CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Modern scientific collaborations have opened up the opportunity of solving complex problems that involve multi-disciplinary expertise and large-scale computational experiments. These experiments usually involve large amounts of data that are located in distributed data repositories running various software systems, and managed by different organisations. A common strategy to make the experiments more manageable is executing the processing steps as a workflow. In this paper, we look into the implementation of fine-grained data-flow between computational elements in a scientific workflow as streams. We model the distributed computation as a directed acyclic graph where the nodes represent the processing elements that incrementally implement specific subtasks. The processing elements are connected in a pipelined streaming manner, which allows task executions to overlap. We further optimise the execution by splitting pipelines across processes and by introducing extra parallel streams. We identify performance metrics and design a measurement tool to evaluate each enactment. We conducted experiments to evaluate our optimisation strategies with a real world problem in the Life Sciences---EURExpress-II. The paper presents our distributed data-handling model, the optimisation and instrumentation strategies and the evaluation experiments. We demonstrate linear speed up and argue that this use of data-streaming to enable both overlapped pipeline and parallelised enactment is a generally applicable optimisation strategy.

The electronics design industry is facing major challenges as transistors continue to decrease in size. The next generation of devices will be so small that the position of individual atoms will affect their behaviour. This will cause the transistors on a chip to have highly variable characteristics, which in turn will impact circuit and system design tools. The EPSRC project "Meeting the Design Challenges of Nano-CMOS Electronics" (Nano-CMOS) has been funded to explore this area. In this paper, we describe the distributed data-management and computing framework under development within Nano-CMOS. A key aspect of this framework is the need for robust and reliable security mechanisms that support distributed electronics design groups who wish to collaborate by sharing designs, simulations, workflows, datasets and computation resources. This paper presents the system design, and an early prototype of the project which has been useful in helping us to understand the benefits of such a grid infrastructure. In particular, we also present two typical use cases: user authentication, and execution of large-scale device simulations.