AI, Volume 6, Issue 12 (December 2025) – 30 articles

Background: Automatic vehicle recognition has recently become an area of great interest, providing substantial support for multiple use cases, including law enforcement and surveillance applications. In real traffic conditions, where for various reasons license plate recognition is impossible or license plates are forged, alternative solutions are required to support human personnel in identifying vehicles used for illegal activities. In such cases, appearance-based approaches relying on vehicle make and model recognition (VMMR) and vehicle color recognition (VCR) can successfully complement license plate recognition. Methods: This research addresses appearance-based vehicle identification, in which VMMR and VCR rely on inherent visual cues such as body contours, stylistic details, and exterior color. In the first stage, vehicles passing through an intersection are detected, and essential visual characteristics are extracted for the two recognition tasks. The proposed system employs deep learning with semantic segmentation and data augmentation for color recognition, while histogram of oriented gradients (HOG) feature extraction combined with a support vector machine (SVM) classifier is used for make-model recognition. For the VCR task, five different neural network architectures are evaluated to identify the most effective solution. Results: The proposed system achieves an overall accuracy of 94.89% for vehicle make and model recognition. For vehicle color recognition, the best-performing models obtain a Top-1 accuracy of 94.17% and a Top-2 accuracy of 98.41%, demonstrating strong robustness under real-world traffic conditions. Conclusions: The experimental results show that the proposed automatic vehicle recognition system provides an efficient and reliable solution for appearance-based vehicle identification. By combining region-tailored data, segmentation-guided processing, and complementary recognition strategies, the system effectively supports real-world surveillance and law-enforcement scenarios where license plate recognition alone is insufficient. Full article

Identifying abnormal group behavior formed by multi-type participants from large-scale historical industry and tax data is important for regulators to prevent potential criminal activity. We propose an Abnormal Alliance detection framework comprising two methods. For detecting joint behavior among multi-type participants, we present DyCIAComDet, a dynamic community identification and tracking method for large-scale, time-varying bipartite multi-type participant networks, and introduce three community-splitting measurement indicators—cohesion, integration, and leadership—to improve community division. To verify whether joint behavior is abnormal, termed an Abnormal Alliance, we propose BMPS, a frequent-sequence identification algorithm that mines key features along community evolution paths based on bitmap matrices, sequence matrices, prefix-projection matrices, and repeated-projection matrices. The framework is designed to address sampling limitations, temporal issues, and subjectivity that hinder traditional analyses and to remain scalable to large datasets. Experiments on the Southern Women benchmark and a real tax dataset show DyCIAComDet yields a mean modularity Q improvement of 24.6% over traditional community detection algorithms. Compared with PrefixSpan, BMPS improves mean time and space efficiency by up to 34.8% and 35.3%, respectively. Together, DyCIAComDet and BMPS constitute an effective, scalable detection pipeline for identifying abnormal alliances in tax datasets and supporting regulatory analysis. Full article

Background: Vision-language models show promise in medical image interpretation, but their performance in musculoskeletal tumor diagnostics remains underexplored. Objective: To evaluate the diagnostic accuracy of six large language models on orthopedic radiographs for tumor detection, classification, anatomical localization, and X-ray view interpretation, and to assess the impact of demographic context and self-reported certainty. Methods: We retrospectively evaluated six VLMs on 3746 expert-annotated orthopedic radiographs from the Bone Tumor X-ray Radiograph dataset. Each image was analyzed by all models with and without patient age and sex using a standardized prompting scheme across four predefined tasks. Results: Over 48,000 predictions were analyzed. Tumor detection accuracy ranged from 59.9–73.5%, with the Gemini Ensemble achieving the highest F1 score (0.723) and recall (0.822). Benign/malignant classification reached up to 85.2% accuracy; tumor type identification 24.6–55.7%; body region identification 97.4%; and view classification 82.8%. Demographic data improved tumor detection accuracy (+1.8%, p < 0.001) but had no significant effect on other tasks. Certainty scores were weakly correlated with correctness, with Gemini Pro highest (r = 0.089). Conclusion: VLMs show strong potential for basic musculoskeletal radiograph interpretation without task-specific training but remain less accurate than specialized deep learning models for complex classification. Limited calibration, interpretability, and contextual reasoning must be addressed before clinical use. This is the first systematic assessment of image-based diagnosis and self-assessment in LLMs using a real-world radiology dataset. Full article

The integration of artificial intelligence (AI) into intelligent control systems has advanced significantly, enabling improved adaptability, robustness, and performance in nonlinear and uncertain environments. This study conducts a PRISMA-2020-compliant systematic mapping of 188 peer-reviewed articles published between 2000 and 15 January 2025, identified through fully documented Boolean queries across IEEE Xplore, ScienceDirect, SpringerLink, Wiley, and Google Scholar. The screening process applied predefined inclusion–exclusion criteria, deduplication rules, and dual independent review, yielding an inter-rater agreement of κ = 0.87. The resulting synthesis reveals three dominant research directions: (i) control model strategies (36.2%), (ii) parameter optimization methods (45.2%), and (iii) adaptability mechanisms (18.6%). The most frequently adopted approaches include fuzzy logic structures, hybrid neuro-fuzzy controllers, artificial neural networks, evolutionary and swarm-based metaheuristics, model predictive control, and emerging deep reinforcement learning frameworks. Although many studies report enhanced accuracy, disturbance rejection, and energy efficiency, the analysis identifies persistent limitations, including overreliance on simulations, inconsistent reporting of hyperparameters, limited real-world validation, and heterogeneous evaluation criteria. This review consolidates current AI-enabled control technologies, compares methodological trade-offs, and highlights application-specific outcomes across renewable energy, robotics, agriculture, and industrial processes. It also delineates key research gaps related to reproducibility, scalability, computational constraints, and the need for standardized experimental benchmarks. The results aim to provide a rigorous and reproducible foundation for guiding future research and the development of next-generation intelligent control systems. Full article

Background: Many everyday electronic devices incorporate embedded computers, allowing them to offer advanced functions such as Internet connectivity or the execution of artificial intelligence algorithms, giving rise to Tiny Machine Learning (TinyML) and Edge AI applications. In these contexts, models must be both efficient and explainable, especially when they are intended for systems that must be understood, interpreted, validated, or certified by humans in contrast to other approaches that are less interpretable. Among these algorithms, linguistic fuzzy systems have traditionally been valued for their interpretability and their ability to represent uncertainty with low computational cost, making them a relevant choice for embedded intelligence. However, in dynamic and changing environments, it is essential that these models can continuously adapt. While there are fuzzy approaches capable of adapting to changing conditions, few studies explicitly address their adaptation and optimization in resource-constrained devices. Methods: This paper focuses on this challenge and presents a lightweight evolutionary strategy, based on a micro genetic algorithm, adapted for constrained hardware online on-device tuning of linguistic (Mamdani-type) fuzzy models, while preserving their interpretability. Results: A prototype implementation on an embedded platform demonstrates the feasibility of the approach and highlights its potential to bring explainable self-adaptation to TinyML and Edge AI scenarios. Conclusions: The main contribution lies in showing how an appropriate integration of carefully chosen tuning mechanisms and model structure enables efficient on-device adaptation under severe resource constraints, making continuous linguistic adjustment feasible within TinyML systems. Full article

Alzheimer’s disease (AD) affects over 50 million people worldwide, making early and accurate diagnosis essential for effective treatment and care planning. Diagnosing AD through neuroimaging continues to face challenges, including reliance on subjective clinical evaluations, the need for manual feature extraction, and limited generalisability across diverse populations. Recent advances in deep learning, especially convolutional neural networks (CNNs) and vision transformers, have improved diagnostic performance, but many models still depend on large labelled datasets and high computational resources. This study introduces an attention-enhanced CNN with a multi-activation fusion (MAF) module and evaluates it using the Alzheimer’s Disease Neuroimaging Initiative dataset. The channel attention mechanism helps the model focus on the most important brain regions in 3D MRI scans, while the MAF module, inspired by multi-head attention, uses parallel fully connected layers with different activation functions to capture varied and complementary feature patterns. This design improves feature representation and increases robustness across heterogeneous patient groups. The proposed model achieved 92.1% accuracy and 0.99 AUC, with precision, recall, and F1-scores of 91.3%, 89.3%, and 92%, respectively. Ten-fold cross-validation confirmed its reliability, showing consistent performance with 91.23% accuracy, 0.93 AUC, 90.29% precision, and 88.30% recall. Comparative analysis also shows that the model outperforms several state-of-the-art deep learning approaches for AD classification. Overall, these findings highlight the potential of combining attention mechanisms with multi-activation modules to improve automated AD diagnosis and enhance diagnostic reliability. Full article

Background: With the development of wireless communication technologies, the rapid advancement of 5G and 6G communication systems has spawned an urgent demand for low latency and high data rates. Orthogonal Frequency Division Multiplexing (OFDM) communication using high-order digital modulation has become a key technology due to its characteristics, such as high reliability, high data rate, and low latency, and has been widely applied in various fields. As a component of cognitive radios, automatic modulation classification (AMC) plays an important role in remote sensing and electromagnetic spectrum sensing. However, under current complex channel conditions, there are issues such as low signal-to-noise ratio (SNR), Doppler frequency shift, and multipath propagation. Methods: Coupled with the inherent problem of indistinct characteristics in high-order modulation, these currently make it difficult for AMC to focus on OFDM and high-order digital modulation. Existing methods are mainly based on a single model-driven approach or data-driven approach. The Adaptive Wavelet Mamba Network (AWMN) proposed in this paper attempts to combine model-driven adaptive wavelet transform feature extraction with the Mamba deep learning architecture. A module based on the lifting wavelet scheme effectively captures discriminative time–frequency features using learnable operations. Meanwhile, a Mamba network constructed based on the State Space Model (SSM) can capture long-term temporal dependencies. This network realizes a combination of model-driven and data-driven methods. Results: Tests conducted on public datasets and a custom-built real-time received OFDM dataset show that the proposed AWMN achieves a performance reaching higher accuracies of 62.39%, 64.50%, and 74.95% on the public Rml2016(a) and Rml2016(b) datasets and our formulated EVAS dataset, while maintaining a compact parameter size of 0.44 M. Conclusions: These results highlight its potential for improving the automatic modulation classification of high-order OFDM modulation in 5G/6G systems. Full article

Background: Artificial intelligence (AI) is increasingly shaping medical education through adaptive learning systems, simulations, and large language models. These tools can enhance knowledge retention, clinical reasoning, and feedback, while raising concerns related to equity, bias, and institutional readiness. Methods: This narrative review examined AI applications in medical and health-profession education. A structured search of PubMed, Scopus, and Web of Science (2010–October 2025), supplemented by grey literature, identified empirical studies, reviews, and policy documents addressing AI-supported instruction, simulation, communication, procedural skills, assessment, or faculty development. Non-educational clinical AI studies were excluded. Results: AI facilitates personalized and interactive learning, improving clinical reasoning, communication practice, and simulation-based training. However, linguistic bias in Natural language processing (NLP) tools may disadvantage non-native English speakers, and limited digital infrastructure hinders adoption in rural or low-resource settings. When designed inclusively, AI can amplify accessibility for learners with disabilities. Faculty and students commonly report low confidence and infrequent use of AI tools, yet most support structured training to build competence. Conclusions: AI can shift medical education toward more adaptive, learner-centered models. Effective adoption requires addressing bias, ensuring equitable access, strengthening infrastructure, and supporting faculty development. Clear governance policies are essential for safe and ethical integration. Full article

Digital images play an increasingly central role in journalism, legal investigations, and cybersecurity. However, modern editing tools make image manipulation difficult to detect with traditional forensic methods. This research addresses the challenge of improving the accuracy and stability of deep-learning-based forgery detection by developing a convolutional neural network enhanced through automated hyperparameter optimisation. The framework integrates a Firefly-based search strategy to optimise key network settings such as learning rate, filter size, depth, dropout, and batch configuration, reducing reliance on manual tuning and the risk of suboptimal model performance. The model is trained and evaluated on a large raster dataset of tampered and authentic images, as well as a custom vector-based dataset containing manipulations involving geometric distortion, object removal, and gradient editing. The Firefly-optimised model achieves higher accuracy, faster convergence, and improved robustness than baseline networks and traditional machine-learning classifiers. Cross-domain evaluation demonstrates that these gains extend across both raster and vector image types, even when vector files are rasterised for deep-learning analysis. The findings highlight the value of metaheuristic optimisation for enhancing the reliability of deep forensic systems and underscore the potential of combining deep learning with nature-inspired search methods to support more trustworthy image authentication in real-world environments. Full article

Deep learning models are increasingly evaluated not only for predictive accuracy but also for their robustness, interpretability, and data quality dependencies. However, current benchmarks largely isolate these dimensions, lacking a unified evaluation protocol that integrates data-centric and model-centric properties. To bridge the gap between data quality assessment and eXplainable Artificial Intelligence (XAI), we introduce DERI1000—the Dataset Explainability Readiness Index—a benchmark that quantifies how suitable and well-prepared a dataset is for explainable and trustworthy deep learning. DERI1000 combines eleven measurable factors—sharpness, noise artifacts, exposure, resolution, duplicates, diversity, separation, imbalance, label noise proxy, XAI overlay, and XAI stability—into a single normalized score calibrated around a reference baseline of 1000. Using five MedMNIST datasets (PathMNIST, ChestMNIST, BloodMNIST, OCTMNIST, OrganCMNIST) and five convolutional neural architectures (DenseNet121, ResNet50, ResNet18, VGG16, EfficientNet-B0), we fitted factor weights through multi-dataset impact analysis. The results indicate that imbalance (0.3319), separation (0.1377), and label noise proxy (0.2161) are the dominant contributors to explainability readiness. Experiments demonstrate that DERI1000 effectively distinguishes models with superficially high accuracy (ACC) but poor interpretability or robustness. The framework thereby enables cross-domain, reproducible evaluation of model performance and data quality under unified metrics. We conclude that DERI1000 provides a scalable, interpretable, and extensible foundation for benchmarking deep learning systems across both data-centric and explainability-driven dimensions. Full article

Objectives: This paper presents an innovative control framework for the classical Cart–Pole problem. Methods: The proposed framework combines Interval Type-2 Fuzzy Logic, the Dueling Double DQN deep reinforcement learning algorithm, and adaptive reward shaping techniques. Specifically, fuzzy logic acts as an a priori knowledge layer that incorporates measurement uncertainty in both angle and angular velocity, allowing the controller to generate adaptive actions dynamically. Simultaneously, the deep Q-network is responsible for learning the optimal policy. To ensure stability, the Double DQN mechanism successfully alleviates the overestimation bias commonly observed in value-based reinforcement learning. An accelerated convergence mechanism is achieved through a multi-component reward shaping function that prioritizes angle stability and survival. Results: Given the training results, the method stabilizes rapidly; it achieves a 100% success rate by episode 20 and maintains consistent high rewards (650–700) throughout training. While Standard DQN and other baselines take 100+ episodes to become reliable, our method converges in about 20 episodes (4–5 times faster). It is observed that in comparison with advanced baselines like C51 or PER, the proposed method is about 15–20% better in final performance. We also found that PPO and QR-DQN surprisingly struggle on this task, highlighting the need for stability mechanisms. Conclusions: The proposed approach provides a practical solution that balances exploration with safety through the integration of fuzzy logic and deep reinforcement learning. This rapid convergence is particularly important for real-world applications where data collection is expensive, achieving stable performance much faster than existing methods without requiring complex theoretical guarantees. Full article

Background: Deep learning models have the potential to enable fast and consistent interpretations of scoliosis radiographs. This study aims to assess the impact of deep learning assistance on the speed and accuracy of clinicians in measuring major coronal curves on scoliosis radiographs. Methods: We utilized a deep learning model (Context Axial Reverse Attention Network, or CaraNet) to assist in measuring Cobb’s angles on scoliosis radiographs in a simulated clinical setting. Four trainee radiologists with no prior experience and four trainee orthopedists with four to six months of prior experience analyzed the radiographs retrospectively, both with and without deep learning assistance, using a six-week washout period. We recorded the interpretation time and mean angle differences, with a consultant spine surgeon providing the reference standard. The dataset consisted of 640 radiographs from 640 scoliosis patients, aged 10–18 years; we divided the dataset into 75% for training, 16% for validation, and 9% for testing. Results: Deep learning assistance achieved non-statistically significant improvements in mean accuracy of 0.32 for trainee orthopedists (95% CI −1.4 to 0.8, p > 0.05) and 0.43 degrees (95% CI −1.6 to 0.8, p > 0.05) for trainee radiologists (non-inferior across all readers). Mean interpretation time decreased by 13.25 s for trainee radiologists, but increased by 3.85 s for trainee orthopedists (p = 0.005). Conclusions: Deep learning assistance for measuring Cobb’s angles was as accurate as unaided interpretation and slightly improved measurement accuracy. It increased the interpretation speeds of trainee radiologists but slightly slowed trainee orthopedists, suggesting that its effect on speed depended on prior experience. Full article

The representation of 3D shapes from point clouds remains a fundamental challenge in computer vision. A common approach decomposes 3D objects into interpretable geometric primitives, enabling compact, structured, and efficient representations. Building upon prior frameworks, this study introduces an enhanced unsupervised deep learning approach for 3D shape representation using superquadrics. The proposed framework fits a set of superquadric primitives to 3D objects through a fully integrated, differentiable pipeline that enables efficient optimization and parameter learning, directly extracting geometric structure from 3D point clouds without requiring ground-truth segmentation labels. This work introduces three key advancements that substantially improve representation quality, interpretability, and evaluation rigor: (1) A uniform sampling strategy that enhances training stability compared with random sampling used in earlier models; (2) An overlapping loss that penalizes intersections between primitives, reducing redundancy and improving reconstruction coherence; and (3) A novel evaluation framework comprising Primitive Accuracy, Structural Accuracy, and Overlapping Percentage metrics. This new metric design transitions from point-based to structure-aware assessment, enabling fairer and more interpretable comparison across primitive-based models. Comprehensive evaluations on benchmark 3D shape datasets demonstrate that the proposed modifications yield coherent, compact, and semantically consistent shape representations, establishing a robust foundation for interpretable and quantitative evaluation in primitive-based 3D reconstruction. Full article

Breast cancer is the second leading cause of cancer-related mortality among women. Early detection enables timely treatment, improving survival outcomes. This paper presents a comparative evaluation of federated learning (FL) frameworks for multiclass breast cancer classification using ultrasound images drawn from three datasets: BUSI, BUS-UCLM, and BCMID, which include 600, 38, and 323 patients, respectively. Five state-of-the-art networks were tested, with MobileNet, ResNet and InceptionNet identified as the most effective for FL deployment. Two aggregation strategies, FedAvg and FedProx, were assessed under varying levels of data heterogeneity in two and three client settings. Results from experiments indicate that the FL models outperformed local and centralized training, bypassing the adverse impacts of data isolation and domain shift. In the two-client federations, FL achieving up to 8% higher accuracy and almost 6% higher macro-F1 scores on average that local and centralized training. FedProx on MobileNet maintained a stable performance in the three-client federation with best average accuracy of 73.31%, and macro-F1 of 67.3% despite stronger heterogeneity. Consequently, these results suggest that the proposed multiclass model has the potential to support clinical workflows by assisting in automated risk stratification. If deployed, such a system could allow radiologists to prioritize high-risk patients more effectively. The findings emphasize the potential of federated learning as a scalable, privacy-preserving infrastructure for collaborative medical imaging and breast cancer diagnosis. Full article

Urban congestion causes further increases in travel times, fuel consumption and greenhouse-gas emissions. In this regard, we conduct a systematic study of a Genetic Algorithm (GA) for real-time routing in an urban scenario in Bethlehem City, based on a SUMO microsimulation that has been calibrated using real data from the field. Our work makes four main contributions: (i) the implementation of a reproducible GA framework for dynamic routing with explicit constraints and adaptive termination criterion; (ii) design of a weight sensitivity study for studying a multi term fitness function with travel time and waiting time, and optionally fuel usage; (iii) an edge-assisted distributed architecture on roadside units (RSUs) supported by cloud services; and (iv) specifying and refining the data set description and experimental protocol with a planned statistical analysis. Empirical evidence from the Bethlehem case study shows a consistent decline in total travel time under high congestion cases. Variations in the waiting time between different scenarios are exhibited, reflecting the trade-offs in the fitness weighting scheme. We recognize that we have some limitations, including the manual resolution of data and the inherent problem of differences between simulations and real world, and we are proposing a road-map towards a pilot deployment that handles these issues. Rather than proposing a new GA variant, we present a deployment-oriented framework-an edge- assisted GA with explicit protocols and a latency envelope, and a reproducible multi-objective tuning procedure validated on a city-scale network under severe congestion. Full article

Occupational safety in high-risk sectors, such as mining, depends heavily on understanding and predicting workers’ behavioural risks. However, existing approaches often overlook the psychological dimension of safety, particularly how psychological-contract violations (PCV) between miners and their organizations contribute to unsafe behavior, and they rarely leverage interpretable artificial intelligence. This study bridges that gap by developing an explainable AutoML framework that integrates AutoGluon, SHAP, and LIME to classify miners’ safety behaviors using psychological and organizational indicators. An empirically calibrated synthetic dataset of 5000 miner profiles (20 features) was used to train multiclass (Safe, Moderate, and Unsafe) and binary (Safe and Unsafe) classifiers. The WeightedEnsemble_L2 model achieved the best performance, with 97.6% accuracy (multiclass) and 98.3% accuracy (binary). Across tasks, Post-Intervention Score, Fatigue Level, and Supervisor Support consistently emerge as high-impact features. SHAP summarizes global importance patterns, while LIME provides per-case rationale, enabling auditable, actionable guidance for safety managers. We outline ethics and deployment considerations (human-in-the-loop review, transparency, bias checks) and discuss transfer to real-world logs as future work. Results suggest that interpretable AutoML can bridge behavioural safety theory and operational decision-making by producing high-accuracy predictions with transparent attributions, informing targeted interventions to reduce unsafe behaviours in high-risk mining contexts. Full article

The development of AI systems and benchmarks has been rapidly increasing, yet there has been a disproportionately small amount of examination into the domains used to evaluate these systems. Most benchmarks introduce bias by focusing on a particular type of domain or combine different domains without consideration of their relative complexity or similarity. We propose the Artificial Intelligence Quotient (AIQ) framework as a means for measuring the similarity and complexity of domains in order to remove these biases and assess the scope of intelligent capabilities evaluated by a benchmark composed of multiple domains. These measures are evaluated with several intuitive experiments using simple domains with known complexities and similarities. We construct test suites using the AIQ framework and evaluate them using known AI systems to validate that AIQ-based benchmarks capture an agent’s intelligence. Full article

Electroencephalography (EEG) provides excellent temporal resolution for brain activity analysis but limited spatial resolution at the sensors, making source unmixing essential. Our objective is to enable accurate brain activity analysis from EEG by providing a fast, calibration-free alternative to independent component analysis (ICA) that preserves ICA-like component interpretability for real-time and large-scale use. We introduce a convolutional neural network (CNN) that estimates ICA-like component activations and scalp topographies directly from short, preprocessed EEG epochs, enabling real-time and large-scale analysis. EEG data were acquired from 44 participants during a 40-min lecture on image processing and preprocessed using standard EEGLAB procedures. The CNN was trained to estimate ICA-like components and evaluated against ICA using waveform morphology, spectral characteristics, and scalp topographies. We term the approach “adaptive” because, at test time, it is calibration-free and remains robust to user/session variability, device/montage perturbations, and within-session drift via per-epoch normalization and automated channel quality masking. No online weight updates are performed; robustness arises from these inference-time mechanisms and multi-subject training. The proposed method achieved an average F1-score of 94.9%, precision of 92.9%, recall of 97.2%, and overall accuracy of 93.2%. Moreover, mean processing time per subject was reduced from 332.73 s with ICA to 4.86 s using the CNN, a ~68× improvement. While our primary endpoint is ICA-like decomposition fidelity (waveform, spectral, and scalp-map agreement), the clean/artifact classification metrics are reported only as a downstream utility check confirming that the CNN-ICA outputs remain practically useful for routine quality control. These results show that CNN-based EEG decomposition provides a practical and accurate alternative to ICA, delivering substantial computational gains while preserving signal fidelity and making ICA-like decomposition feasible for real-time and large-scale brain activity analysis in clinical, educational, and research contexts. Full article

Helicobacter pylori infects about half of the global population and is a major cause of peptic ulcer disease and gastric cancer. Improving patient education can increase screening participation, enhance treatment adherence, and help reduce gastric cancer incidence. Recently, large language models (LLMs) such as ChatGPT, Gemini, and DeepSeek-R1 have been explored as tools for producing patient-facing educational materials; however, their performance compared to expert gastroenterologists remains under evaluation. This narrative review analyzed seven peer-reviewed studies (2024–2025) assessing LLMs’ ability to answer H. pylori-related questions or generate educational content, evaluated against physician- and patient-rated benchmarks across six domains: accuracy, completeness, readability, comprehension, safety, and user satisfaction. LLMs demonstrated high accuracy, with mean accuracies typically ranging from approximately 77% to 95% across different models and studies, and with most models achieving values above 90%, comparable to or exceeding that of general gastroenterologists and approaching senior specialist levels. However, their responses were often judged as incomplete, described as “correct but insufficient.” Readability exceeded the recommended sixth-grade level, though comprehension remained acceptable. Occasional inaccuracies in treatment advice raised safety concerns. Experts and medical trainees rated LLM outputs positively, while patients found them less clear and helpful. Overall, LLMs demonstrate strong potential to provide accurate and scalable H. pylori education for patients; however, heterogeneity between LLM versions (e.g., GPT-3.5, GPT-4, GPT-4o, and various proprietary or open-source architectures) and prompting strategies results in variable performance across studies. Enhancing completeness, simplifying language, and ensuring clinical safety are key to their effective integration into gastroenterology patient education. Full article

This self-study presents Persona, Break Glass, Name Plan, Jam (PBNJ), a human-centered workflow for using generative AI to support differentiated lesson planning and problem solving. Although differentiated instruction (DI) is widely endorsed, early-career teachers often lack the time and capacity to implement it consistently. Through four iterative cycles of collaborative self-study, seven educator-researchers examined how they used AI for lesson planning, identified key challenges, and refined their approach. When engaged, the PBNJ sequence—set a persona, use a ‘break glass’ starter prompt, name a preliminary plan, and iteratively ‘jam’ with the AI—improved teacher confidence, yielded more feasible lesson plans, and supported professional learning. We discuss implications for problem solving beyond educational contexts and the potential for use with young learners. Full article

Artificial Intelligence (AI) is increasingly framed as a driver of economic transformation, yet its capacity to alleviate poverty in the Global South remains contested. This article introduces the notion of AI Economics—the political economy of value creation, extraction, and redistribution in AI systems—to interrogate h ow innovation agendas intersect with structural inequalities. This article examines how Social Innovation (SI) systems, when coupled with decentralized Web3 technologies such as blockchain, Decentralized Autonomous Organizations (DAOs), and data cooperatives, may challenge data monopolies, redistribute economic gains, and support inclusive development. Drawing on Action Research (AR) conducted during the AI4SI International Summer School in Donostia-San Sebastián, this article compares two contrasting ecosystems: (i) the Established AI4SI Ecosystem, marked by centralized governance and uneven benefits, and (ii) the Decentralized Web3 Emerging Ecosystem, which promotes community-driven innovation, data sovereignty, and alternative economic models. Findings underscore AI’s dual economic role: while it can expand digital justice, service provision, and empowerment, it also risks reinforcing dependency and inequality where infrastructures and governance remain weak. This article concludes that embedding AI Economics in context-sensitive, decentralized social innovation systems—aligned with ethical governance and the SDGs—is essential for realizing AI’s promise of poverty alleviation in the Global South. Full article

Background: Degenerative cervical spondylosis (DCS) is a common and potentially debilitating condition, with surgery indicated in selected patients. Deep learning models (DLMs) can improve consistency in grading DCS neural stenosis on magnetic resonance imaging (MRI), though existing models focus on axial images, and comparisons are mostly limited to radiologists. Methods: We developed an enhanced transformer-based DLM that trains on sagittal images and optimizes axial and foraminal classification using a maximized dataset. DLM training utilized 648 scans, with internal testing on 75 scans and external testing on an independent 75-scan dataset. Performance of the DLM, spine surgeons, and radiologists of varying subspecialities/seniority were compared against a consensus reference standard. Results: On internal testing, the DLM achieved high agreement for all-class classification: axial spinal canal κ = 0.80 (95%CI: 0.72–0.82), sagittal spinal canal κ = 0.83 (95%CI: 0.81–0.85), and neural foramina κ = 0.81 (95%CI: 0.77–0.84). In comparison, human readers demonstrated lower levels of agreement (κ = 0.60–0.80). External testing showed modestly degraded model performance (κ = 0.68–0.77). Conclusions: These results demonstrate the utility of transformer-based DLMs in multiplanar MRI interpretation, surpassing spine surgeons and radiologists on internal testing and highlighting its potential for real-world clinical adoption. Full article

Artificial intelligence is reshaping healthcare project management in Saudi Arabia, yet most deployments lack culturally grounded ethics. This paper synthesises global AI-ethics guidance and Islamic bioethics, then proposes a maqāṣid-al-sharīʿah-aligned conceptual framework for ANN-based decision support. Ethical signals derived from the preservation of life, dignity, justice, faith, and intellect are embedded as logic-gate filters on ANN outputs. The framework specifies a dual-metric evaluation that reports predictive performance (e.g., accuracy, MAE, AUC) alongside ethical compliance, with auditable thresholds for fairness (δ = 0.1) and confidence (α = 0.8) calibrated through stakeholder workshops. It incorporates a co-design protocol with clinicians, patients, Islamic scholars, and policymakers to ensure cultural and clinical legitimacy. Unlike UNESCO and EU frameworks, which remain principle-oriented, this study introduces a measurable dual-layer assessment that combines technical accuracy with ethical compliance, supported by audit artefacts such as model cards, traceability logs, and human override records. The framework yields technically efficient and Shariah-compliant recommendations and sets a roadmap for empirical pilots under Vision 2030. The paper moves beyond a general review by formalising an Islamic-values-driven conceptual framework that operationalises ethical constraints inside ANN–DSS pipelines and defines auditable compliance metrics. This paper combines a critical review of AI in healthcare project management with the development of a maqāṣid-aligned conceptual framework, thereby bridging systematic synthesis with an implementable proposal for ethical AI. Full article

Objectives: This study aims to present an AI-driven threat-hunting framework that automates both hypothesis generation and hypothesis validation through a hybrid deep learning model that combines Convolutional Neural Networks (CNN) with Long Short-Term Memory (LSTM) networks. The objective is to operationalize proactive threat hunting by embedding anomaly detection within a structured workflow, improving detection performance, reducing analyst workload, and strengthening overall security posture. Methods: The framework begins with automated hypothesis generation, in which the model analyzes network flows, telemetry data, and logs sourced from IoT/IIoT devices, Windows/Linux systems, and interconnected environments represented in the TON_IoT dataset. Deviations from baseline behavior are detected as potential threat indicators, and hypotheses are prioritized according to anomaly confidence scores derived from output probabilities. Validation is conducted through iterative classification, where CNN-extracted spatial features and LSTM-captured temporal features are jointly used to confirm or refute hypotheses, minimizing manual data pivoting and contextual enrichment. Principal Component Analysis (PCA) and Recursive Feature Elimination with Random Forest (RFE-RF) are employed to extract and rank features based on predictive importance. Results: The hybrid model, trained on the TON_IoT dataset, achieved strong performance metrics: 99.60% accuracy, 99.71% precision, 99.32% recall, an AUC of 99%, and a 99.58% F1-score. These results outperform baseline models such as Random Forest and Autoencoder. By integrating spatial and temporal feature extraction, the model effectively identifies anomalies with minimal false positives and false negatives, while the automation of the hypothesis lifecycle significantly reduces analyst workload. Conclusions: Automating threat-hunting processes through hybrid deep learning shifts organizations from reactive to proactive defense. The proposed framework improves threat visibility, accelerates response times, and enhances overall security posture. The findings offer valuable insights for researchers, practitioners, and policymakers seeking to advance AI adoption in threat intelligence and enterprise security. Full article

Brain tumors are abnormal tissue growth characterized by uncontrolled and rapid cell proliferation. Early detection of brain tumors is critical for improving patient outcomes, and magnetic resonance imaging (MRI) has become the most widely used modality for diagnosis due to its superior image quality and non-invasive nature. Deep learning, a subset of artificial intelligence, has revolutionized automated medical image analysis by enabling highly accurate and efficient classification tasks. The objective of this study is to develop a robust and effective brain tumor detection system using MRI images through transfer learning. A diagnostic framework is constructed based on convolutional neural networks (CNN), integrating both a custom sequential CNN model and pretrained architectures, namely VGG16 and EfficientNetB4, trained on the ImageNet dataset. Prior to model training, image preprocessing techniques are applied to enhance feature extraction and overall model performance. This research addresses the common challenge of limited MRI datasets by combining EfficientNetB4 with targeted preprocessing, data augmentation, and an appropriate optimizer selection strategy. The proposed methodology significantly reduces overfitting, improves classification accuracy on small datasets, and remains computationally efficient. Unlike previous studies that focus solely on CNN or VGG16 architectures, this work systematically compares multiple transfer learning models and demonstrates the superiority of EfficientNetB4. Experimental results on the Br35H dataset show that EfficientNetB4, combined with the ADAM optimizer, achieves outstanding performance with an accuracy of 99.66%, precision of 99.68%, and an F1-score of 100%. The findings confirm that integrating EfficientNetB4 with dataset-specific preprocessing and transfer learning provides a highly accurate and cost-effective solution for brain tumor classification, facilitating rapid and reliable medical diagnosis. Full article

In capital-intensive industries, operational knowledge is often trapped in unstructured technical manuals, creating a barrier to efficient and reliable maintenance planning. This work addresses the need for an integrated system that can automate knowledge extraction and generate optimized, resilient, operational plans. A synergistic multi-agent framework is introduced that transforms unstructured documents into a structured knowledge base using a self-validating pipeline. This validated knowledge feeds a scheduling engine that combines multi-objective optimization with discrete-event simulation to generate robust, capacity-aware plans. The framework was validated on a complex maritime case study. The system successfully constructed a high-fidelity knowledge base from unstructured manuals and the scheduling engine produced a viable, capacity-aware operational plan for 118 interventions. The optimized plan respected all daily (6) and weekly (28) task limits, executing 64 tasks on their nominal date, bringing 8 forward, and deferring 46 by an average of only 2.0 days (95th percentile 4.8 days) to smooth the workload and avoid bottlenecks. An interactive user interface with a chatbot and planning calendar provides verifiable “plan-to-page” traceability, demonstrating a novel, end-to-end synthesis of document intelligence, agentic AI, and simulation to unlock strategic value from legacy documentation in high-stakes environments. Full article

This paper examines how Generative AI (GenAI) reshapes digital inclusion in e-government. We develop the E-Government Divide Measurement Indicator (EGDMI) across three dimensions: D1—Breadth of the Divide (foundational access, affordability, and basic skills), D2—Sectoral/Specific Divide (actual use, experience, and trust in e-government), and D3—GenAI Gap (access, task use, and competence). The index architecture specifies indicator lists, sources, units, transformations, uniform normalization, and a documented weighting strategy with sensitivity and basic uncertainty checks. Using official statistics and qualitative evidence for Serbia, we report D1 and D2 as composite indices and treat D3 as an exploratory, non-aggregated layer given current data maturity. Results show strong foundational readiness (D1 = 73.6) but very low e-government uptake (D2 = 19.9), indicating a shift of the divide from access to meaningful use, usability, and trust. GenAI capabilities are emergent and uneven (D3 sub-dimensions: access 47.8; task use 39.4; competence/verification 43.6). Cluster analysis identifies four user profiles—from “Digitally Excluded” to “GenAI-Augmented Citizens”— that support differentiated interventions. The initial hypothesis—that GenAI can widen disparities in the short run—receives partial confirmation: GenAI may lower interaction costs but raises verification and ethics thresholds for vulnerable groups. We outline a policy roadmap prioritizing human-centered service redesign, transparency, and GenAI literacy before automation, and provide reporting templates to support comparable monitoring and cross-country learning. Full article

Accurate trajectory learning and reconstruction represent a core challenge in mobile robotics, particularly in environments affected by sensor noise, drift, and incomplete data. Addressing this challenge is essential for reliable navigation and motion control in real-world Internet of Robotic Things (IoRT) systems. This paper presents a data-driven framework for learning and reconstructing mobile robot trajectories using LSTM autoencoders. Trajectory data were collected from both simulation and real-world experiments with a Unitree GO1 quadruped robot, preprocessed through normalization, sequence padding, and trajectory boundary flags, and then used to train recurrent neural network models. The proposed architecture employs bidirectional LSTM layers and a custom loss function combining reconstruction, velocity, and boundary terms to improve trajectory stability. Experimental results show stable reconstruction accuracy across simulated and real-world datasets, with the position RMSE reduced from 0.92 m to 0.60 m and the yaw MAE improved from 0.49 rad to 0.17 rad on the most complex trajectory. The evaluation was conducted in controlled indoor conditions and offline mode, which defines the current scope of validation. Future work will extend the analysis to larger and more diverse environments and investigate extensions such as attention mechanisms, sensor fusion, and online learning to enhance adaptability in real-world deployment. Full article

Background: Diagnostic errors can be substantially diminished, and clinical decision-making can be significantly enhanced through automated image classification. Methods: We implemented a YOLO (You Only Look Once)-based system to classify diabetic retinopathy (DR) utilizing a unique retinal dataset. Although YOLO provides exceptional accuracy and rapidity in object recognition and categorization, its interpretability is constrained. Both binary and multi-class classification methods (graded severity levels) were employed. The Contrast-Limited Adaptive Histogram Equalization (CLAHE) model was utilized to improve image brightness and detailed readability. To improve interpretability, we utilized Eigen Class Activation Mapping (Eigen-CAM) to display areas affecting classification predictions. Results: Our model exhibited robust and consistent performance on the datasets for binary and 5-class tasks. The YOLO 11l model obtained a binary classification accuracy of 97.02% and an Area Under Curve (AUC) score of 0.98. The YOLO 8x model showed superior performance in 5-class classification, with an accuracy of 80.12% and an AUC score of 0.88. A simple interface was created using Gradio to enable real-time interaction. Conclusions: The suggested technique integrates robust prediction accuracy with visual interpretability, rendering it a potential instrument for DR screening in clinical environments. Full article

Background: The accelerating global population aging underscores the urgency of addressing public health challenges. Sarcopenia and depression are prevalent, interrelated conditions in older adults, yet prevailing research often treats depression as a static state, neglecting its longitudinal progression and limiting predictive capability for sarcopenia. Methods: Using data from four waves (2011–2018) of the China Health and Retirement Longitudinal Study (CHARLS), we identified distinct depressive symptom trajectories via Group-Based Trajectory Modeling. Seven machine learning algorithms were employed to develop predictive models for sarcopenia risk, incorporating these trajectory patterns and baseline characteristics. Results: Three depressive symptom trajectories were identified: ‘Persistently Low’, ‘Persistently Moderate’, and ‘Persistently High’. Tree-based ensemble methods, particularly Random Forest and XGBoost, demonstrated superior and robust performance (mean accuracy: 0.8265 and 0.8178; mean weighted F1-score: 0.8075 and 0.8084, respectively). Feature importance analysis confirmed depressive symptoms as a core, independent predictor, ranking third (5.7% importance) in the optimal Random Forest model, only after BMI and cognitive function, and surpassing traditional risk factors like age and waist circumference. Conclusions: This study validates that longitudinal depressive symptom trajectories provide superior predictive power for sarcopenia risk compared to single-time-point assessments, effectively mapping mental health trajectories to physical risk. The robust ML framework not only enables early identification of high-risk individuals but also reveals a multidimensional risk profile, highlighting the intricate mind–body connection in aging. These findings advocate for integrating dynamic mental health monitoring into routine geriatric assessments, demonstrating the potential of AI to facilitate a paradigm shift towards proactive, personalized, and scalable prevention strategies in public health and clinical practice. Full article