Interpretable machine learning and deep learning model for discriminating pheochromocytoma from adrenocortical adenoma based on CT: A multicenter study.

Deep Learning Algorithm Based on Contrast-Enhanced Ultrasound Potentially Optimizes Treatment Strategies for Solitary Primary Hepatocellular Carcinoma.

Field-scale potato yield prediction from sentinel-2 time series using lightweight deep learning models

With the rise of the Internet of Medical Things (IoMT), healthcare systems increasingly rely on Wireless Body Area Networks (WBANs) for continuous, real-time patient monitoring and clinical decision-making. These applications require ultra-low latency, high reliability, and energy efficiency. Typically, they operate via mobile devices, such as smartphones, wearables, or WBAN coordinators, which collect, process, and transmit medical data. However, the limited processing capabilities and energy constraints of these devices often lead to increased delays and degraded system performance. To address these challenges, Mobile Edge Computing (MEC) has emerged as a promising solution that brings computation closer to the network edge. This paper addresses the optimization problem of task offloading and resource allocation in WBAN-MEC systems, where each task can be executed locally on the mobile device, offloaded to the MEC server, or to the cloud. The problem is formulated as a Mixed-Integer Nonlinear Programming (MINLP) model involving offloading decisions and the allocation of communication and computational resources. Our objective is to maximize task completion subject to time constraints, minimize mobile energy consumption, and ensure efficient use of MEC resources. We propose a Collaborative Multi-Agent Task Offloading and Resource Allocation (CoMA-TORA) framework, which decomposes the complex optimization problem into two coordinated components: a decentralized offloading decision component and a centralized resource allocation component. The framework is implemented using an actor-critic reinforcement learning architecture, with a global critic that evaluates a shared reward for coordinated decision-making. Simulation results show that CoMA-TORA outperforms both traditional and DRL-based approaches in delay-sensitive healthcare environments.

Arrhythmia is a condition in which a person's heartbeat is irregular and can pose serious health risks. Effective arrhythmia detection is necessary to reduce associated risks. This study aims to develop a new deep learning architecture combining Convolutional Neural Networks (CNNs) to extract features, Long Short-Term Memory (LSTMs) to handle sequential data, and Gated Recurrent Units (GRUs) to reduce computational resources, leveraging the strengths of each to achieve better classification accuracy for diagnosing using the MIT-BIH Arrhythmia Database. The data is preprocessed by 0.5 Hz (low) and 50 Hz (high) to remove noise, then segmented into smaller, normalized signals to a unique scale, determine the higher point of the QRS complex in ECG, then gets labeled each segment, and finally converted ECG segment to (2D). The proposed model outperforms models on CNN, LSTM, and GRU if we apply them alone, with a precision of 92%, F1-score of 97%, recall of 100%, and accuracy of 97%; this study's notable discovery is that the suggested method may substantially decrease the duration when using RNN networks in conjunction with CNN. This paper presents a cost-effective approach to ECG signal reduction and a robust automatic scheme for arrhythmia detection, leveraging the strengths of CNN, LSTM, and GRU networks. The suggested model has achieved significant improvements in accuracy and is potentially a useful tool for real-time clinical practice. • A novel hybrid deep learning model combining CNN, LSTM, and GRU for ECG analysis. • Robust preprocessing pipeline improves ECG signal quality and arrhythmia classification. • Achieves 97% accuracy, 100% recall, and 97% F1-score on the MIT-BIH Arrhythmia dataset. • Outperforms standalone CNN, LSTM, and GRU models in efficiency and accuracy. • Demonstrates potential for real-time, cost-effective arrhythmia diagnosis in healthcare.

Comparative analysis of traditional and deep learning time series architectures for influenza A infectious disease forecasting.

Hydro-environmental dynamics of Kaptai Lake using satellite derived biophysical metrics and an ensemble Machine Learning Framework.

Artificial intelligence for marine oil spill management: Recent advances and future directions.

Challenges in translating AI-driven ASD/ADHD diagnosis: A methodological systematic review.

Classification of familial and non-familial ADHD using auto-encoding network and binary hypothesis testing.

A dual-branch temporal-spatial deep learning architecture for efficient and accurate CAN bus intrusion detection

Seafood provides high-quality protein and essential nutrients but is highly susceptible to rapid postharvest deterioration. Conventional quality evaluation methods are often destructive, labor-intensive, and difficult to implement in real-time industrial and consumer settings. In recent years, deep learning-assisted computer vision (DL-CV) has emerged as a promising technical route for nondestructive and rapid seafood quality assessment in industrial processing lines and retail inspection systems. This review synthesizes recent advances in DL-CV for seafood quality evaluation from both technical and application-oriented perspectives. The technical framework is first clarified by comparing visible-light imaging with other imaging modalities and by discussing the transition from traditional machine learning to end-to-end deep learning-based feature extraction. Recent developments in representative deep learning architectures, such as convolutional neural networks and vision transformers, are then summarized alongside emerging trends toward lightweight design, architectural enhancement, and interpretability. Current application scenarios are further reviewed systematically, with particular emphasis on freshness evaluation, including key image regions and two dominant labeling strategies: storage time-based labeling (STBL) and traditional indicator-based labeling (TIBL). Additional applications such as species identification, weight estimation, defect detection, and quantitative determination of compositional attributes are also discussed. Overall, DL-CV demonstrates strong potential for accurate, nondestructive seafood quality prediction, especially when leveraging visible-light imaging systems that are easily deployable in practical environments. Future research should focus on finer quality differentiation, multidimensional and integrated quality evaluation, and scalable deployment in both industrial processing and consumer-oriented applications.

Deep learning-based segmentation of enamel, cementum, alveolar bone, and gingiva in periodontal ultrasound images.

This study evaluated the efficacy of optimized deep learning architectures using a portable fluorescence imaging device specifically for the in-situ detection of fecal contamination on chicken eggshells to enhance food safety. The research utilized a Contamination and Sanitization Inspection device to establish a comprehensive dataset of fluorescence images, leveraging the spectral characteristics of fecal matter which emits fluorescence in the 600 to 720 nm range. Based on this fluorescence image data set, the study developed high performance models to identify fecal residues across both brown and white eggshells. Experimental results demonstrated that the fluorescence signals of fecal contaminants remain highly stable under ambient lighting, with both the primary mode utilizing 405 nm excitation and the enhance mode utilizing 365 nm excitation achieving Structural Similarity Index Measure (SSIM) values consistently exceeding 0.9200. These metrics confirm that the intrinsic high contrast of fluorescence imaging maintains structural integrity without the need for strict darkroom environments. Through the evaluation of nine distinct neural networks, it was found that the 365 nm excitation effectively suppressed background interference on brown eggs, allowing the lightweight MobileNet architecture to detect fecal contamination with an accuracy of 0.9000. For white eggshells, the 405 nm excitation coupled with the ViT Base 384 model yielded a peak accuracy of 0.9333 in identifying minute fecal traces. The reliability of the detection was further validated through Explainable AI frameworks which confirmed that the classification logic was consistently based on actual contaminated regions marked by fecal residues. These findings provide a robust methodology for leveraging handheld portable fluorescence technology to establish objective standards for detecting fecal contamination in the poultry industry.

A novel Contrast Based Learning technique for training Deep Learning architectures on small datasets

Epileptic seizure prediction is a critical research area that enables timely intervention and prevention of severe neurological complications. With the growing integration of IoT in healthcare, real-time EEG monitoring has become essential for continuous and automated seizure detection. The suggested approach presents a hybrid deep learning architecture that integrates various sophisticated computational approaches to deliver precise, safe, and effective seizure prediction. EEG data are recorded in real time with an IoT-based headband and processed with Shape-Aware Mesh Normal Filtering (SMNF) in order to eliminate noise and enhance the quality of the signal. In addition to that, the Quadratic Phase Quaternion Domain Fourier Transform (QPQDFT) is the feature extraction principle that is effective in both spectral and temporal variations. The features extracted are then categorized with Physics-Penalized Dual-Branch Spectral-Spatial Neural Network (PP-DBSSNN), which employs physics-based regularization and dual-branch attention as a way of enhancing generalization and interpretability of the data. Finally, Key Escrow-Free Attribute-Based Encryption (KEF-ABE) is a method that guarantees the security and privacy of EEG information on clouds. The findings of the experiment show the best performance with an accuracy of 99.95%, a precision of 99.93%, and a specificity of 99.91% in the case of the Bonn EEG dataset, and an accuracy of 99.96%, a precision of 99.94%, and a specificity of 99.92% in the case of the CHB-MIT dataset, which confirms its robustness and reliability.

Deep learning architectures for Brugada syndrome detection: A comparative analysis with GAN-generated ECG data

The realistic and controllable generation of pure smoke is critical for smoke image editing, smoke visual special effects generation, and smoke data synthesizing within security scenarios. It is a relatively underexplored topic and continues to present significant challenges. Existing methods face challenges in the generation of smoke with intricate details and the regulation of various smoke styles. In this paper, a Pure Smoke image Generation Network (PSGNet) is proposed with a gradient and style learning approach to generate realistic and controllable smoke images. To achieve flexibility in control across the spatial dimension, the smoke shape mask is used to encode spatial details, such as the location and contour of the smoke, along with other related properties. To enhance the physical realism of synthesized smoke, a novel gradient-based learning framework is proposed to generate smoke gradient features, highlighting a special focus on explicitly encoding and exploiting gradient information. This framework uses a smoke gradient learning architecture that captures the subtle structures and patterns characteristic of real smoke, enabling the generation of highly realistic smoke with rich, fine-scale detail. In addition, a spatially aware style learning strategy is proposed to provide fine-grained control over smoke attributes such as density, color, and overall look. It is able to effectively model style features across both channel and spatial dimensions, thereby enabling spatially aware style manipulation. By combining the gradient module with this style learning framework, the method produces smoke that exhibits rich visual details and customizable image styles. Experiments conducted on six benchmark datasets demonstrate that the proposed PSGNet significantly outperforms the state-of-the-art approaches.

Multimodal artificial intelligence (AI) approaches integrating heterogeneous data sources represent an emerging frontier in liver fibrosis assessment. However, use of multimodal AI for liver fibrosis staging has been only preliminarily explored, and the existing evidence is constrained by substantial methodological gaps. This scoping review aimed to comprehensively map the current evidence on multimodal AI models that integrate medical imaging with other data categories for predicting liver fibrosis stage. Following the Joanna Briggs Institute methodology and PRISMA-ScR guidelines, we searched MEDLINE, Web of Science, CENTRAL, and IEEE Xplore on August 12, 2025. Studies developing AI or machine learning models for liver fibrosis prediction integrating at least one imaging modality with heterogeneous data categories (e.g., clinical parameters or serum biomarkers) were included. Three reviewers independently screened records, and extracted data were independently verified by two additional reviewers. Of 2,849 records, 21 studies met the eligibility criteria, yielding 34 distinct multimodal AI models. Research was geographically concentrated in China (81%) and predominantly focused on hepatitis B-related liver disease. CT-based radiomics combined with serum biomarkers represented the most common approach, whereas deep learning architectures were less frequently applied. Across 107 AUC evaluations, the median AUC was 0.890 (interquartile range 0.850-0.925). External-validation AUCs (12 evaluations from 6 studies) ranged 0.808-0.990; 3 internal-test AUCs from a single study fell below 0.70. However, external validation was reported for only 20.6% of models, with calibration and decision curve analysis reported in 23.1% and 24.1% of evaluations, respectively. This scoping review revealed a nascent field with encouraging diagnostic performance but with substantial gaps in external validation, calibration reporting, and clinical utility assessment. Future research should prioritize methodologically rigorous validation and evaluate the impact on clinical decision-making.

AW-EL-PINNs: A multi-task learning physics-informed neural network for Euler-Lagrange systems in optimal control problems.