Cracks pose a significant threat to road and building safety, making effective detection of cracks on road surfaces a focus of research both domestically and internationally. Deep learning-based methods often require extensive pixel-level annotations, posing significant labor costs. We propose a single-stage weakly supervised crack segmentation model based on multi-scale feature fusion. The model is built on a single-stage weakly supervised segmentation framework, which reduces model complexity. It utilizes a multi-scale feature fusion module (PPM) to integrate features at different scales, enhancing the ability to extract features from cracks of various sizes. The combination of the Domain Restriction Suppression (DRS) module and pixel affinity convolution is employed to optimize pseudo-pixel annotations. In addition, we propose a joint loss function to mitigate sample imbalance between crack and non-crack pixels. Compared to other two-stage weakly supervised segmentation models, our model is simpler and more effective, achieving excellent results on the Deep Crack and Crack500 datasets, surpassing most weakly supervised crack segmentation models in terms of Recall (Re), F-score (F1), and mean Intersection-over-Union (mIoU), achieving similar effects to fully supervised crack segmentation models. This demonstrates the effectiveness and robustness of our model.

QRS complex detection is a crucial preprocessing step in ECG-based arrhythmia recognition. Traditional methods often fail to maintain stability when confronted with noisy interference, baseline drift, and high inter-patient variability in ECG signals. Although deep learning approaches achieve superior detection performance, their high computational complexity hinders deployment on mobile and edge devices. Therefore, this study proposes a lightweight encoder-decoder network named MSCRAG for robust QRS complex detection. The architecture integrates joint time and frequency analysis with multi-scale convolutional to enable adaptive frequency decomposition, and incorporates a channel attention to enhance QRS time and frequency feature representation. In the encoder and decoder, an improved GhostV2Block (GhostV2BlockMS) reduces model parameters and computational cost while preserving feature extraction efficiency. Furthermore, the Multi-Scale Dynamic Feature Convolution Plus (MSDFC+) module which introduces in the bottleneck enhances the global receptive field and provides dynamic feature modulation, allowing the network to adaptively focus on critical temporal regions and capture long-range dependencies in ECG signals. Evaluations on standard ECG datasets demonstrate that our method achieves 99.1 % accuracy, 97.2 % sensitivity, 99.6 % positive predictive value, and 99.3% F1-score, outperforming existing state-of-the-art approaches. Its high accuracy, lightweight design, and low computational cost render it suitable for both static analysis and real-time cardiac monitoring in dynamic and wearable applications, offering significant potential for clinical and portable healthcare scenarios.

Neoadjuvant chemoradiotherapy (nCRT) is the standard treatment for locally advanced rectal cancer (LARC). However, substantial inter-patient variability in response to nCRT poses a significant challenge for accurately predicting treatment outcomes based on preoperative data, thereby complicating clinical decision-making. With the rapid advancement of artificial intelligence technologies, there has been growing interest in leveraging AI for predictive modeling in cancer therapy. In this study, we propose a novel multi-modal prediction framework, FMPNet, which integrates preoperative magnetic resonance imaging (MRI) and whole-slide image (WSI) features to predict nCRT efficacy. Specifically, for WSI processing, we develop an efficient tumor cell segmentation strategy and incorporate a deep subspace clustering mechanism into the feature extraction pipeline to enhance the model's representational capacity. Comprehensive experiments demonstrate that FMPNet consistently outperforms ten other feature fusion-based prediction models on both internal test sets and external validation cohorts across multiple metrics, including accuracy, precision, recall, F1-score and ROC-AUC curves. These results not only confirm the superior performance of our model but also underscore its potential to support more accurate and personalized clinical decision-making for patients with LARC.

Automated quality assessment of cardiac ultrasound images is crucial for ensuring diagnostic accuracy and clinical decision-making reliability. Hospitals face significant challenges in efficiently screening ultrasound image quality, where manual expert review is time-consuming and subjective. However, dedicated methods for ultrasound image quality assessment remain scarce, with most adapted from natural image quality metrics that fail to capture clinically relevant diagnostic factors. In this paper, we propose a novel dual-domain framework that models both spatial anatomical features and frequency-domain spectral characteristics using specialized neural modules. Our approach incorporates cardiac-specific attention mechanisms and wavelet-based artifact detection to enable comprehensive, clinically aligned evaluation. Extensive experiments on a largescale clinical dataset demonstrate the superiority of our method, achieving 88.1 % overall accuracy, a macro-averaged F1-score of 88.1 %, and 100 % precision and recall in detecting diagnostically unacceptable images. The proposed framework is designed to meet clinical reliability standards, paving the way for safe integration into diagnostic workflows.

Metal artifacts severely degrade the quality of CT images. Existing learning-based metal artifact reduction (MAR) methods often miss multi-scale anatomy and fail to transfer from synthetic to clinical scans. We present CLEAR-Net, which injects clinical priors and adaptively aligns corrupted features via two modules: CEAB, a quantized multi-scale anatomy bank distilled from clean clinical CTs, and FLAG, a scale-wise gating mechanism that aligns features to CEAB priors across domains. This structure-aware design preserves organ boundaries and fine textures, boosting robustness and generalization. Extensive experiments demonstrate CLEAR-Net's superior performance. The source code will be made publicly available.

The power system of electrically propelled aircraft is trending towards the development of hybrid energy forms. Different types of energy sources have distinct characteristics, and the coordination of hybrid energy sources can improve the performance of the power system. The hybrid energy form studied in this paper consists of a fuel cell and a lithium battery. Considering the unique operational conditions of power systems, this study, based on rule-based energy management strategies, proposes a dynamic balance energy management strategy based on the hydrogen consumption of the fuel cell. This strategy ensures that the energy consumption of the fuel cell and the auxiliary power source remains relatively balanced, preventing the depletion of one power source before the other. It is capable of adapting to various operating conditions, enhancing the energy utilization and stability of the hybrid power system, and ensuring the reliability of the UAV's power system. Simulation results confirm the feasibility of the proposed strategy. Finally, the hardware for the energy management system is designed and experimentally validated. The experimental results and subsequent analysis confirm the viability of the proposed energy management strategy.

Although weakly-supervised techniques can reduce the labeling effort, it is unclear whether a saliency model trained with weakly-supervised data (e.g., point annotation) can achieve the equivalent performance of its fully-supervised version. This paper attempts to answer this unexplored question by proving a hypothesis: there is a point-labeled dataset where saliency models trained on it can achieve equivalent performance when trained on the densely annotated dataset. To prove this conjecture, we proposed a novel yet effective adversarial spatio-temporal ensemble active learning. Our contributions are four-fold: 1) Our proposed adversarial attack triggering uncertainty can conquer the overconfidence of existing active learning methods and accurately locate these uncertain pixels. 2) Our proposed spatio-temporal ensemble strategy not only achieves outstanding performance but significantly reduces the model's computational cost. 3) Our proposed relationship-aware diversity sampling can conquer oversampling while boosting model performance. 4) We provide theoretical proof for the existence of such a point-labeled dataset. Experimental results show that our approach can find such a point-labeled dataset, where a saliency model trained on it obtained 98%-99% performance of its fully-supervised version with only ten annotated points per image.

The forward problem of electrocardiography is essential for understanding the mechanisms underlying cardiac electrical activity, validating and optimizing the inverse problem, and advancing personalized medicine through patient-specific modeling. Comprehensive computation of the forward problem requires multi-scale modeling and the execution of complex computational processes. The challenges associated with constructing, solving, and validating multi-scale models arise from the need to accurately represent biophysical processes by establishing precise interactions across various spatial and temporal scales. In this study, we aimed to integrate single-cell modeling, which captures the electrophysiological activity of cardiac cells, with finite element methods (FEM), which simulates the propagation of electrical signals from cardiac tissue to the body surface, to achieve a comprehensive simulation of the electrocardiogram (ECG) forward problem. We developed a multi-scale tissue structure model, incorporating it with single-cell models and FEM to compute the distribution of body surface potentials. Additionally, we simulated a 12-lead ECG by appropriately positioning electrodes. The results demonstrate that our proposed approach effectively simulates the QRS and T-wave components of the ECG and accurately captures variations in sub-cellular parameters under both normal and abnormal conditions as reflected in the ECG.

This paper presents a comprehensive evaluation of the performance of different numerical methods in cardiac electrophysiological simulations, including the Explicit Euler (EE) Method, Implicit Euler (IE) Method, Trapezoidal Rule (TR), and the Fourth-Order Runge-Kutta (RK4) Method within the framework of one-dimensional (1D), two-dimensional (2D), and three-dimensional (3D) models of the human ventricles. Simulation accuracy was investigated for varying the time step (h) for different tissue scales and complexities. It was shown that, as the dimension of the model increased, the upper limit of h for acceptable simulation accuracy (hul) gradually decreased. However, the difference in hulamong the methods was small. As the h or the tissue dimension increased, the EE method showed greater computational error compared to other methods. The experiment results also showed that the amplitude of the external stimulation pulse current, the size of the spatial range of tissue receiving stimuli, the spatial dimension of the tissue models, as well as the diffusion coefficient of the model, also affected simulation accuracy. The present study discusses the influence of numerical methods on the accuracy of cardiac simulations, providing insights for choosing optimal numerical methods for cardiac simulations.

The incidence of ventricular tachycardia has been steadily increasing year by year. Current clinical interventions primarily involve medication and radiofrequency ablation surgery. However, due to technical limitations, precise and effective localization of ablation sites remains challenging, leading to prolonged surgery times and impacting patient outcomes. In this study, we propose using diffusion models to establish a bidirectional mapping relationship between ECG and cardiac electrophysiology, unifying the forward and inverse problem-solving processes of cardiac electrophysiological signals. Specifically, during the training phase, we add Gaussian noise to cardiac electrophysiological signals through the forward process, progressively constructing the distribution of ECG and Gaussian mixed noise. Subsequently, the reverse process is utilized to denoise the ECG and Gaussian mixed noise, learning the distribution of cardiac electrophysiological data. Ultimately, the trained diffusion model establishes a mapping relationship from ECG to cardiac electrophysiology, thus achieving the inverse problem-solving of the electrophysiological model.

Medical image registration (MIR) is essential for various clinical diagnoses and treatments. Despite the rapid progress in deep learning-based MIR techniques, most methods focus on directly optimizing the raw image intensity information. In this paper, we explore the usage of edge information of anatomical structures associated with the spatial location of image intensities to assist in registration, termed EdgeReg. The intuition is that the edge information can provide additional rich boundary information to the raw images, enhancing the network’s feature representation. Additionally, as the edge images are strictly spatially consistent with the raw images, additional supervised information can be added to network training. Specifically, we first extract the edge images from the raw moving and fixed images using the Sobel operator and feed these images into a lightweight feature extractor to merge the image intensity and edge information. The enriched features are subsequently input into established registration networks. Finally, similarity loss is applied to both the raw and edge images. Extensive experiments show that EdgeReg is compatible with various networks across diverse datasets and dimensions (2D and 3D), achieving superior registration performance. In particular, EdgeReg does not rely on segmentation labels and is trained in an unsupervised paradigm. Therefore, edge information is a beneficial assistance for unsupervised MIR. The code is available at https://github.com/PerceptionComputingLab/EdgeReg.

Abstract Neural Radiance Fields (NeRFs) have achieved significant recognition for their proficiency in scene reconstruction and rendering by utilizing neural networks to depict intricate volumetric environments. Despite considerable research dedicated to reconstructing physical scenes, rare works succeed in challenging scenarios involving dynamic, multi‐material objects. To alleviate, we introduce CoupNeRF, an efficient neural network architecture that is aware of multiple material properties. This architecture combines physically grounded continuum mechanics with NeRF, facilitating the identification of motion systems across a wide range of physical coupling scenarios. We first reconstruct specific‐material of objects within 3D physical fields to learn material parameters. Then, we develop a method to model the neighbouring particles, enhancing the learning process specifically in regions where material transitions occur. The effectiveness of CoupNeRF is demonstrated through extensive experiments, showcasing its proficiency in accurately coupling and identifying the behavior of complex physical scenes that span multiple physics domains.

Arsenic trioxide $(As2\mathrm{O}_{3}$), an antileukemia drug, has been used to treat acute promyelocytic leukemia (APL) for more than fifty years, and its therapeutic effect has been elucidated at the molecular level. However, several side effects were observed in APL patients administrated with $As2\mathrm{O}_{3}$, such as long QT (LQT) syndrome, torsade de pointes tachycardia, and even sudden cardiac death. This means that the clinically relevant dosage may induce severe cardiotoxicity. Accordingly, it is essential to determine the underlying mechanisms of arrhythmia induced by $\mathrm{As}2\mathrm{O}_{3}$. Some biological experiments indicated that $\mathrm{As}2\mathrm{O}_{3}$ can impair the human ether-à-go-gorelated gene (hERG), thus inhibiting rapid delayed rectifier potassium current $(I_{Kr})$ and prolonging action potential duration (APD), which was regarded as the reason for LQT syndrome. However, previous experiments did not illuminate the deep mechanisms of $\mathrm{As}2\mathrm{O}_{3}$-induced side effects, which is important in clinical treatment. In addition, the experimental data were restricted to animal studies, so human cellular data were lacking. In this study, we investigated $\mathrm{As}2\mathrm{O}_{3}$-related cardiotoxicity through a human ventricular model study. Based on the current experimental data, the effects of $\mathrm{As}2\mathrm{O}_{3}$ on ventricular myocytes (VMs) were predicted at various $\mathrm{As}2\mathrm{O}_{3}$ concentrations. In addition, the potential hazard of $\mathrm{As}2\mathrm{O}_{3}$ was simulated and illustrated under different stimulation protocols. Moreover, electrocardiograms (ECGs) were estimated in heterogeneous ventricular cables, by which the clinical phenomenon was verified and explained. Based on the present modeling study, deep reasons for arrhythmia caused by $\mathrm{As}2\mathrm{O}_{3}$ were uncovered. $\mathrm{As}2\mathrm{O}_{3}$ not only led to a prolonged APD but also alternated action potentials and exacerbated heterogeneity among VMs. Moreover, the degree of arrhythmia risk was susceptible to $\mathrm{As}2\mathrm{O}_{3}$ dosage. These new findings may provide targets for attenuating $\mathrm{As}2\mathrm{O}_{3}$ toxicity and may help to improve the APL therapeutic regimen.

Biological pacemaker is a therapy for cardiac rhythm disease, which can be transformed from ventricular myocytes (VMs) by overexpressing HCN gene which codes the expression of hyperpolarization-activated current (${\mathrm {I}}_{\mathrm{f}}$) and knocking off Kir2.1 gene which codes inward-rectifier potassium current (${\mathrm {I}}_{\mathrm{K1}}$). Our previous study built a biological pacemaker single cell model and clarified the underlying mechanisms of how gene expressing levels influence the pacemaking activity of single pacemaker cell. But the pacemaking ability of pacemaker tissue has not been researched systematically. And what factors may have effects on pacemaker’s synchronization and spontaneous beating propagation are not clear. Biological research indicated that both sinoatrial node and pacemaker cells has less expression of connexin than unexcitable cardiac cells, which provides a possibility that improve pacemaking ability of pacemaker by decreasing its cell coupling. Another possible factor is the number of pacemaker cells. According to the common sense, increasing cell number can promote pacemaking behaviours, but overmuch pacemaker cells is unreasonable in clinic. As a result, the balance between pacemaker number and cell coupling is important when applying biological pacemaker. In this study, we constructed a two-dimensional cardiac tissue model with the description of electrophysiology to illustrate the relationship between gap junction and cell number. Based on this model, we modified the cell coupling between pacemaker cells by adjusting the diffusion coefficient of tissue with different pacemaker number. In different condition, the synchronization, pacemaking cycle length and electrical signal propagation were evaluated. It can be concluded that weakening cell coupling among pacemaker cells can lift the efficiency of bio-pacemaker therapy. This study may contribute to produce effective pacemaker in clinic.

Pacemaking dysfunction (PD) may result in heart rhythm disorders, syncope or even death. Current treatment of PD using implanted electronic pacemakers has some limitations, such as finite battery life and the risk of repeated surgery. As such, the biological pacemaker has been proposed as a potential alternative to the electronic pacemaker for PD treatment. Experimentally and computationally, it has been shown that bio-engineered pacemaker cells can be generated from non-rhythmic ventricular myocytes (VMs) by knocking out genes related to the inward rectifier potassium channel current (IK1) or by overexpressing hyperpolarization-activated cyclic nucleotide gated channel genes responsible for the "funny" current (If). However, it is unclear if a bio-engineered pacemaker based on the modification of IK1- and If-related channels simultaneously would enhance the ability and stability of bio-engineered pacemaking action potentials. In this study, the possible mechanism(s) responsible for VMs to generate spontaneous pacemaking activity by regulating IK1 and If density were investigated by a computational approach. Our results showed that there was a reciprocal interaction between IK1 and If in ventricular pacemaker model. The effect of IK1 depression on generating ventricular pacemaker was mono-phasic while that of If augmentation was bi-phasic. A moderate increase of If promoted pacemaking activity but excessive increase of If resulted in a slowdown in the pacemaking rate and even an unstable pacemaking state. The dedicated interplay between IK1 and If in generating stable pacemaking and dysrhythmias was evaluated. Finally, a theoretical analysis in the IK1/If parameter space for generating pacemaking action potentials in different states was provided. In conclusion, to the best of our knowledge, this study provides a wide theoretical insight into understandings for generating stable and robust pacemaker cells from non-pacemaking VMs by the interplay of IK1 and If, which may be helpful in designing engineered biological pacemakers for application purposes.

The biological pacemaker was a promising therapy for cardiac diseases such as sick sinus syndrome and atrioventricular block. A lot of experiments showed that pacemaker cells can be transformed from non-rhythmic cardiac cells or stem cells by gene therapy. However, at the tissue level, the electrophysiological properties between rhythmic and non-rhythmic regions are different. For example, the expression of connexin (such as Cx43) reduced in the induced-pacemaker cells which means that the pacemaker cells may have a less electrical coupling with adjacent cells. In addition, some researches indicated that the spatial distribution of pacemaker cells influenced the excitability of cardiac tissue. To the best of our knowledge, it is still unclear how the spatial distribution of bio-pacemaker cells affects the pacemaking behaviour in biological pacemaker tissue. In this study, we constructed a series of two-dimensional pacemaker-ventricle models containing different distributions of pacemaker cells to investigate the effect of spatial distribution on the pacemaking behaviour. Three kinds of models were designed in our simulations: (1) Tight model; (2) Embedded model; (3) Electrically isolated model. The pacemaking ability of cardiac tissue was measured by the least ratio of pacemaker cells needed to drive the whole tissue. Simulation results showed that electrically isolated model was the optimal model as it showed the best pacemaking ability among these three models. This study may guide the clinical use of bio-pacemaker.

The biological pacemaker was a promising therapy for cardiac diseases such as sick sinus syndrome and atrioventricular block. A lot of experiments showed that pacemaker cells can be transformed from non-rhythmic cardiac cells or stem cells by gene therapy. However, at the tissue level, the electrophysiological properties between rhythmic and non-rhythmic regions are different. For example, the expression of connexin (such as Cx43) reduced in the induced-pacemaker cells which means that the pacemaker cells may have a less electrical coupling with adjacent cells. In addition, some researches indicated that the spatial distribution of pacemaker cells influenced the excitability of cardiac tissue. To the best of our knowledge, it is still unclear how the spatial distribution of bio-pacemaker cells affects the pacemaking behaviour in biological pacemaker tissue. In this study, we constructed a series of two-dimensional pacemakerventricle models containing different distributions of pacemaker cells to investigate the effect of spatial distribution on the pacemaking behaviour. Three kinds of models were designed in our simulations: (1) Tight model; (2) Embedded model; (3) Electrically isolated model. The pacemaking ability of cardiac tissue was measured by the least ratio of pacemaker cells needed to drive the whole tissue. Simulation results showed that electrically isolated model was the optimal model as it showed the best pacemaking ability among these three models. This study may guide the clinical use of biopacemaker.

It has been verified that biological pacemaker could be produced based on ventricular myocytes (VMs) by overexpressing HCN gene which codes the expression of hyperpolarization-activated current (If). Clinically, xenograft is in common use by which one specie' stem cell is infected with another specie's HCN gene so that the stem cell could transfer into cardiac pacemaker cell. The difference of HCN gene between species affects Ifproperties, but how the Ifproperties influence pacemaker creation is not easy to be qualified in biological experiments. In this study, we build an engineered biological pacemaker model based on a ventricular myocyte model by incorporating Ifformulation and simulated the membrane potential of biological pacemaker. The Ifof different species is simulated by modifying average half-maximal activation voltage (V1/2) of Ifactivation gate and Ifconductance (Gf). Based on the modified pacemaker model, the effect of Ifproperties on pacemaking stability and frequency is evaluated. Simulation results indicate that pacemaking ability is influenced dramatically by Ifproperties. In addition, the spontaneous pacemaking mechanism showed both membrane-clock and Ca2+-clock and its deep reason is analyzed in this paper. This study may provide a subcellular perspective for the clinical use of biological pacemaker.