Accurate brain tumor segmentation from magnetic resonance imaging (MRI) is essential for diagnosis, treatment planning, and disease monitoring in neuro-oncology. Although convolutional neural network (CNN)-based methods, such as U-Net, have achieved success, they often struggle to preserve fine structural details and accurately delineate tumors with irregular boundaries, particularly when tumors are small or have complex morphology. Based on this motivation, this paper proposes a novel framework that integrates topological features into a U-Net-based segmentation model. Persistent homology is applied to 2D MRI images to extract topological cycles that capture morphological and structural characteristics of brain tissue including tumors. These features are transformed into feature maps and incorporated as an additional input channel along with the original MRI images, providing complementary topological information to CNN. The proposed framework is evaluated on the brain tumor dataset using standard segmentation metrics, Mean Intersection over Union (mIoU) and the segmentation results were validated by a medical expert. Experimental results indicate that the integration of topological features consistently improves segmentation performance compared to the standard U-Net model, particularly in preserving tumor boundaries and fine details. This study highlights the potential of combining topological data analysis with deep learning to enhance robustness and accuracy in medical image segmentation.

Carpal Tunnel Syndrome (CTS) is a prevalent condition that requires precise assessment of hand and wrist functionality for early diagnosis and effective treatment. Traditional diagnostic methods often necessitate in-clinic visits, which can be time-consuming and costly for both patients and healthcare providers. This study proposes a novel approach by utilizing a mobile phone app to evaluate potential CTS patients’ hand and wrist movements through a task as simple as lifting a virtual martini glass. The app leverages the phone's accelerometer and gyroscope sensors to collect precise motion data, offering a portable and modern alternative to conventional methods. This approach is motivated by the accessibility and ubiquity of smartphones, making it an innovative way to democratize healthcare tools. It holds the potential to reduce the strain on clinical resources and eliminate barriers for patients who may face difficulties traveling to healthcare facilities. Moreover, this method underscores the importance of integrating digital technologies into healthcare, paving the way for more efficient, cost-effective, and patient-centered diagnostic processes.

YOLO is a cutting-edge object detection model that has demonstrated exceptional effectiveness in object recognition. However, YOLO is sensitivity to similar-shaped objects, leading to false positive detections. Our experimental results show that YOLO mistakenly identified non-sticker objects with high confidence due to their shape resemblance to parking stickers. Increasing the confidence threshold lowered these false positives but also caused real stickers to be missed. Persistent homology provides a topological approach to image analysis by representing image domains as simplicial complexes. In this study, we applied persistent homology features to enhance parking sticker identification in vehicle images, addressing the limitations of traditional deep-learning-based detection methods. We employed YOLO-V10 for initial sticker detection. However, misclassifications occurred due to the similarity of certain objects to stickers, prompting us to integrate persistent homology features, such as critical pairs, persistence values, and bounding box areas, into machine learning classifiers. Among several tested models, the HistGradient Boosting classifier exhibited the best performance, which significantly improving classification accuracy. Experimental results demonstrated that persistent homology features provide a robust solution for distinguishing parking stickers from visually similar objects, effectively complementing YOLO-V10’s detection capabilities.

Artificial Intelligence (AI) and machine learning (ML) are increasingly integral to decision-making processes in various aspects of our lives. However, their predictions could be biased against protected social groups. Existing bias mitigation techniques often require the manual selection of protected attributes, limiting their generalizability. In this paper, we present AutoFairMask, a model-based extrapolation method for the bias mitigation with the automatic protected attributes detection, comparing three approaches: (1) a base model with no fairness interventions, (2) a base FairMask model with a single manually selected protected attribute, and (3) AutoFairMask, an enhanced FairMask model with the automatic protected attribute detection. Our experimental results show that FairMask with the autodetection performs better at minimizing biases while preserving the predictive performance than both the baseline and manually defined FairMask. Notably, the autodetection approach uncovers additional sources of biases (e.g., police district) that are missed by the manual selection. This emphasizes how crucial adaptive bias reduction strategies are to the development of fairness-aware AI. Our findings suggest that automated identification multi-protected attributes are able to maintain the fairness and accuracy in real-world ML applications.

Respiratory disease presents major global health challenges, and its early detection could save numerous lives, prompting the demand for accurate and non-intrusive diagnostic methods. This study presents a multi-model machine learning framework for classifying respiratory diseases (COPD, Bronchiolitis, Pneumonia, URTI, and Healthy) using audio data from the Respiratory Sound Database. We extract Mel frequency cepstral coefficients (MFCCs) and apply data augmentation to address class imbalance. We propose a novel hybrid model, ConvMambda, and compare its performance against other popular classification models, including 1D CNN, random forest, SVM, and LSTM. The ConvMamba model achieves the highest test accuracy (90.77%) and macro F1-score (0.8932), outperforming 1D-CNN (90% accuracy) and random forest (82% accuracy), SVM (68% accuracy), and LSTM (68% accuracy). We analyze the performance across classes, highlighting challenges with minority classes (Bronchiolitis, URTI) and propose strategies to mitigate imbalance. Our findings demonstrate the potential of audio-based diagnostics, with the ConvMamba model offering the superior performance for respiratory diseases classification.

The extreme weather caused by climate changes has caused a lot of serious damages across the globe. Wildfire is one of the disasters that increasingly threaten the public safety around the world, such as the wildfire in Canada back in 2023, causing the massive ash flow and affected multiple areas around including cities in the neighboring countries. Researchers have developed various simulation models to predict the wildfire occurrence and its spread. However, the prediction accuracy is limited due to the unavailability of the real time parameters, such as the fuel and GIS data of the fire field. With the abundance of satellite image data, deep learning has attracted researchers' attention for the wildfire detection. In this paper, we propose a deep learning method using Convolutional Neural Networks (CNNs) to predict the wildfire based on the satellite image data set. The experimental results show the achieved prediction accuracy reaches up to 96%, validating the effectiveness of the proposed model.

The tissue paper converting system involves critical stages, including unwinding, embossing, folding, stacking, cutting, and packaging. Despite its maturity, challenges such as production interruptions during roll changes, equipment coordination issues, and unstable packaging speeds persist, affecting overall efficiency. While existing methods like fast roll change mechanisms and automated speed regulation systems have improved individual equipment performance, they fall short in achieving system-wide optimization. This paper proposes a digital-twin-based framework for the tissue paper converting system. High-precision simulation models are developed for key equipment, including the base paper roll stand, folding machine, paper stacker, and packaging machine, enabling real-time system monitoring and prediction. Key performance indicators such as base paper availability, folding speed, stacking status, and packaging progress are effectively predicted, preventing interruptions and reducing downtime. The proposed framework was validated in a real-world application at a Chinese paper manufacturing company. Results show that it increases machine runtime by 3 h per day, improving production efficiency by 15%. This leads to an additional daily output of 35,000 packs, contributing approximately CNY 85,000 in revenue growth. Additionally, the system helps prevent 300 kg of base paper waste daily, saving CNY 600 in manufacturing costs per day.

Evolutionary algorithms (EAs) have been widely applied in many engineering optimization problems. However, the choice of an appropriate value of the popular size (PS) is confused. For different applications, a large PS plays a positive or negative role in the performance of EAs. To facilitate the choice of the PS for EAs, we develop an exclusion operator to guide EAs to avoid visiting the non-optima solutions to reduce the sensitivity to the PS. The exclusion operator prevents EAs from revisiting the solutions in a repository, which contains those non-optimal solutions. This method decreases the volume of the valid search space during the evolution process, and therefore lessens EAs’ sensitivity to the PS, thus easing their configurations. Experimental results show that on average the sensitivity of 33 benchmark functions was reduced by 25.10%, and the ranges of suggested PS values are 10-40 and 10-20, in which 90% and 95% of cases are much more insensitive to the PS value than the canonical EA, significantly reduced to its 3.6% and 3.26%, respectively. Furthermore, the proposed work builds the solid foundation for its wide adoption in many other optimization problems, especially those involving expensive experiments.

Reversible data hiding in encrypted images (RDHEI) is a challenging task since it requires the complex processing of encrypted images and the lossless recovery after the data extraction. Several RDHEI methods have been proposed to balance the privacy and security of images in cloud environments. However, they can only embed less identifier information. Recently, a novel encoding technology called parametric binary tree labeling (PBTL) has gained popularity. However, existing PBTL-based schemes either inefficiently encode non-embeddable pixels or leave them unprocessed. In light of this issue, we propose an efficient parametric binary tree labeling method for RDHEI (EPBTL-RDHEI). In the pixel division, we design a novel EPBTL encoding rule to encode all pixels except reference pixels into three groups including non-embeddable pixels, embeddable pixels, and self-recording pixels. In this way, the length of auxiliary information is reduced. Furthermore, we derive theoretical conditions for the improved embedding performance with the introduction of self-recording pixels. Experiments demonstrate that the payload capacity of our presented EPBTL encoding surpasses that of the state-of-the-art RDHEI schemes. The average embedding rates on public datasets such as BOSSbase, BOWS-2, and UCID reach up to 2.667, 2.637, and 2.370 bpp (bit per pixel) with a full reversibility, and the improvement percentages are up to 175.24%, 232.80%, and 1,842.29% compared to the two comparative PBTL-based methods.

Text images have a more uniform background than natural images, making it difficult to capture tampering traces. Thus, it poses significant challenges to the detection and localization of tampered text images. To effectively detect tampered text images, we propose the edge-guided multi-feature fusion network (EMF-Net) including four different modules, the difference semantic discriminator (DSD), the edge-guidance feature aggregation module (EFAM), the edge supervised module (ESM), and the multi-branch attention-induced fusion module (MAFM). The proposed network combines RGB images and learned noise-sensitive features to deeply excavate the hidden tampered region features to improve the detection accuracy due to the insufficient feature information. Meanwhile, the multi-branch attention-induced fusion and semantic discriminator methods are integrated to reduce false alarms by shielding the feature interference in non-tampered regions and diminish the external interference. Furthermore, we craft a dataset of 12,000 text images and their tempered versions with three tempering operations including copy-move, splicing and inpainting. Extensive experiments have shown that the proposed network can improve the generalization performance and achieve the higher detection accuracy compared to current other state-of-the-art methods.

Parallel/Distributed particle filters estimate the states of dynamic systems by using Bayesian interference and stochastic sampling techniques with multiple processing units (PUs). The sampling procedure and the resampling procedure alternatively execute to estimate the states in particle filters. There are two basic types of resampling techniques used in parallel/distributed particle filters. They are centralized resampling and decentralized resampling. The high communication between PUs in centralized resampling lowers the speedup factor in parallel computing but improves the estimation accuracy. The decentralized resampling can avoid the communication and improve the performance. Some types of hybrid resampling techniques mainly execute the decentralized resampling and only invoke the centralized resampling with constant intervals to achieve ideal performance without losing the estimation accuracy. However, the constant intervals cannot guarantee that the centralized resamplings are invoked timely. In this study, we proposed a hybrid resampling technique with adaptive intervals between centralized resamplings to overcome that issue. The experimental results indicate that the proposed hybrid resampling technique is able to improve the performance and the estimation accuracy.

This study explores the potential of Artificial Intelligence (AI) in early screening and prognosis of Dry Eye Disease (DED), aiming to enhance the accuracy of therapeutic approaches for eye-care practitioners. Despite the promising opportunities, challenges such as diverse diagnostic evidence, complex etiology, and interdisciplinary knowledge integration impede the interpretability, reliability, and applicability of AI-based DED detection methods. The research conducts a comprehensive review of datasets, diagnostic evidence, and standards, as well as advanced algorithms in AI-based DED detection over the past five years. The DED diagnostic methods are categorized into three groups based on their relationship with AI techniques: (1) those with ground truth and/or comparable standards, (2) potential AI-based methods with significant advantages, and (3) supplementary methods for AI-based DED detection. The study proposes suggested DED detection standards, the combination of multiple diagnostic evidence, and future research directions to guide further investigations. Ultimately, the research contributes to the advancement of ophthalmic disease detection by providing insights into knowledge foundations, advanced methods, challenges, and potential future perspectives, emphasizing the significant role of AI in both academic and practical aspects of ophthalmology.

In modeling and simulation, face validation with domain experts is a recommended validation technique applied after implementing the conceptual model. Experts can offer early feedback on the prediction results. Experts’ empirical experiences can identify mistakes in the conceptual model and help estimate initial values of key parameters to save time for later calibrations. This work proposes a three-phase method to conduct face validation with police investigators and crime analysts to validate an agent-based crime risk prediction model. The three phases focus on verifying the prediction results, the behavioral design of target individuals, and the risk estimation process respectively. We present each phase’s goals, details, and types of validity addressed to provide clear guidance on adopting the method. Quantitative results of the feedback have shown that our face validation improves the correctness of the conceptual model and boosts intended users’ confidence in the model results. Our work also provides additional empirical experiences in urban crime prediction. The experimental results show the potential of the proposed method for validating other software systems with domain experts’ knowledge.

Based on the rapid development of intelligent technologies in recent years, the digital transformation of the whole industry and society has become increasingly important. Among them, digital twins and artificial intelligence have great potentials in improving industry processes and further enhancing productivity. This paper proposes an Intelligent Digital Twin System (IDTS) based on artificial intelligence and digital twins for the paper industry. The system includes the prediction models for the stirring speed of the dump chest, the water consumption of the deflaker, the supply air pressure of the dryer, and the exhaust air temperature of the dryer. The sensors, 5G network slices, and other equipment collect related data during the papermaking process for generating twin data, and we use the prediction models to analyze the data and monitor important indicators (stirring speeds, water consumptions, supply air pressures, and exhaust air temperatures) for the manufacturing processes, which are used to improve the energy utilization and production efficiency of the paper industry and thus facilitate cost saving. We apply this intelligent digital twin system and its associated prediction models to an actual paper manufacturing factory and show their effectiveness by improving the operational efficiency and saving labor and maintenance costs.

Carpal tunnel syndrome (CTS) is one of the common neurological disorders caused by prolonged compression of the median nerve. Thus, CTS patients’ daily tasks are significantly affected. Traditional diagnostic methods are invasive or subjective, causing pain or inaccuracy. Therefore, a more accurate machine/deep learning classifier is needed to provide an accessible assessment approach that can help screen out early-stage patients to prevent further deterioration. Behavioral biomechanics has shown great potential to be used for CTS and its severity classification. The biomechanical parameters are collected when identified patients and healthy individuals perform daily life activities, such as grasping and lifting in a controlled manner. Facing the challenges of time series biomechanical data with small sample sizes and high dimensions, we propose a novel classification algorithm to create an ensemble model for CTS detection using Long Short-Term Memory (LSTM). The proposed algorithm achieves 93% accuracy on average for CTS detection using biomechanical data of daily life activities.

It requires innovations to accurately predict crimes and offending risks due to the complex nature of human behaviors within different social contexts. In recent years, theory-based and data-driven agent-based models have become useful tools in computational criminology with the abundance of public data, the rise of computing powers, and extensive explorations of related theories and modeling tools. In this paper, we present CARESim, an integrated simulation environment to predict the risks of high-risk individuals committing street violent crimes. The simulation environment combines agent-based modeling with a real geographical information system and incorporates public data (e.g., environmental data, crime data, census data, and transportation data) to repro-duce authentic human mobility. This simulation environment is capable of replicating crime patterns and evaluating policies. We demonstrate the use of this environment and illustrate how key factors, such as social networks, the neighborhood, the weather, police patrol strategies, the population, and citizens' behaviors, impact the risk of street-level violent offending. Furthermore, the environment can support the optimization of patrol strategies and offer evaluations of different social setups that could be unfeasible or expensive to test in real life.

In multipartner projects, team members from two or more organizations often have different skills and behavioral preferences. A major purpose of managing such teams is fostering cooperation among different types of team members, which is critical for the successful completion of a project. Traditional studies assumed that the preferences of members within a team are homogeneous and examined their effects on cooperation. Motivated by the heterogeneity of team members' preferences within a team, we propose a model of effort expenditures in multipartner project teams with two heterogeneous members to explore how the cooperative effort expenditure (CEE) of one agent is affected by the preference type of a coworker. We theoretically formulate the problem, provide the solution, and conduct the parameter sensitivity analysis. Furthermore, we present a numerical example to demonstrate the usage of the model and gain insight into the mechanism of the effect of heterogeneous preferences on the team cooperation. The theoretical analysis and experimental results showed that the effort cost and potential value of collaboration are the most critical factors that determine the interaction among the team member's optimal CEE, project outputs, and the coworker's preference type. The payoff differences between team members play a complex moderating role in the effect of team members' heterogeneity preference types on team cooperation. This paper contributes to the body of knowledge on project management by understanding the cooperative behavior of team members with heterogeneous behavioral preferences, and providing managerial insights for project managers to promote cooperation in multipartner project teams.

High-order problems pose significant challenges for evolutionary algorithms (EAs) to optimize. To mitigate this, a deep hybrid transfer learning EA (DHTL-EA) is proposed. DHTL-EA works by transferring both the model and the optima from a corresponding low-order problem. Here, a deep neural network is adopted to model both the low-order and high-order problems, and the training data are derived from historical evolutionary data. The theoretical basis of DHTL-EA is well-supported by the transferability of solutions among domination-landscape-equivalent problems. The transfer to high-order problem is achieved by retraining the last fully connected layer of the deep neural network model for the target problem. Experiments on two groups of problems validated that DHTL-EA is effective on high-order problems. Further testing on other benchmark functions demonstrated its competitive performance.

This article introduces a mesh-less, ray-based deep neural network method to solve the Helmholtz equation with high frequency. This method does not use an adaptive mesh refinement method, nor does it design a numerical scheme using some specially designed basis function to calculate the numerical solution, but it has the advantages of easy implementation and no mesh. We have carried out various numerical examples to prove the accuracy and efficiency of the proposed numerical method.