Abdulbasit A. Darem

In medical image analysis, transfer learning from ImageNet-pretrained models is a dominant paradigm, but its effectiveness on small clinical datasets is questionable. In this work, we challenge this understanding by showing that paradigms for natural image encoders can fundamentally fail to suit the medical imaging context for early disease detection, where data are scarce. We present MORSE (Morphological Observation with Reweighted Spectral Encoding), a minimal architecture that incorporates specific inductive biases by performing multi-scale morphological operations, adaptive gating of the spectrally decomposed signals, and memory-efficient hierarchical fusion. MORSE is evaluated on four clinical medical imaging tasks from dermatology, ophthalmology and neurology, and shows consistent improvements over eight state-of-the-art (SoTA) baselines trained from random initialization, as well as ImageNet-pretrained models and a RadImageNet-pretrained ResNet-50 (the only RadImageNet backbone tested). With limited training data, MORSE outperforms the strongest from-scratch baselines by 0.83 to 5.00 percentage points and exceeds the best pretrained models by 0.23 to 2.27 percentage points, whereas pure vision transformers underperform by approximately 16 to 45 percentage points on the three smaller datasets. Ablation studies support the complementary contributions of morphological and spectral features, and parameter analysis shows that architecture design using medical imaging priors can be more consequential than model capacity in the evaluated small-data settings. These studies challenge the prevailing reliance on transfer learning in data-constrained medical imaging, suggesting that domain-specific architectural innovation is a viable alternative for specialized diagnostic tasks.

The rapid expansion of the internet of things (IoT) has enabled large‐scale connectivity across healthcare, smart homes, industrial automation, and intelligent infrastructure. However, this growth has also increased the exposure of IoT environments to complex and evolving cyber threats. Traditional intrusion detection systems, particularly signature‐based approaches, are often ineffective against previously unseen attacks and struggle to adapt to the heterogeneous and dynamic nature of IoT traffic. To address these challenges, this study proposes a hybrid intrusion detection framework that combines generative adversarial learning with graph attention‐based modeling. The proposed model leverages adversarial data generation to improve the representation of minority attack classes and employs graph attention mechanisms to capture structural dependencies among communicating entities. The framework was evaluated using the UNSW‐NB15 dataset and compared with baseline deep learning models, including generative adversarial networks, graph convolutional networks, and graph attention networks. The proposed method achieved an accuracy of 81.23%, precision of 83.89%, recall of 78.01%, and F 1‐score of 80.84% on the held‐out test set, while also reducing false‐positive and false‐negative rates relative to the comparison models. The results demonstrate the effectiveness of combining adversarial data augmentation with graph attention‐based representation learning under the controlled, offline evaluation conditions used in this study. Although the framework may be relevant to IoT and industrial cybersecurity applications, its scalability, real‐time performance, edge‐device feasibility, and effectiveness in operational environments require further experimental validation.

Background: The mining sector’s digital transformation increasingly relies on AI-driven digital twins (AI-DTs) that integrate real-time data with intelligent analytics. A recent systematic literature review (SLR) of 68 studies identified a critical gap: which technical choices guarantee industrial success? Objective: This study extends that SLR by validating a Deployment Readiness Score (DRS) to identify which combinations of technical and methodological choices are associated with high readiness AI-DT studies in the literature. Methods: Each study was coded across eight dimensions and assigned to a DRS based on data source, validation method, and operational metric reporting. A random forest classifier was used as a consistency check for the DRS framework. Results: The model achieved 100% test accuracy as an internal consistency check within the coded dataset, confirming that the DRS scoring rules produce a coherent classification across the reviewed studies. The data source was the strongest association (41.2%), followed by publication year (25.6%) and validation method (22.7%). AI technique showed minimal association (0.9%). Studies using sensory data achieved 100% high readiness within the DRS framework; mixed data achieved 95.2%; and experimental validation achieved 95.7%. The proportion of high-readiness studies increased from 35.7% in 2024 to 93.5% in 2025. Conclusions: Within the reviewed literature, high-fidelity data and rigorous validation show stronger associations with high readiness than algorithmic complexity. We provide a Deployment Readiness Scorecard and propose minimal reporting standards, shifting focus from theoretical algorithms to practical data acquisition and validation.

The recognition of user behavioural patterns provides an innovative method for access control systems, as it relies on automatic detection of human behaviour and does not require user involvement or the recall of credentials. These actions can be captured surreptitiously by passive sensors placed within the user’s environment. Existing solutions fail to integrate real-time physiological and behavioural patterns for robust user verification. This work presents an advanced multimodal user behavioural authentication framework that incorporates data from electroencephalogram (EEG) signals, Internet of Things (IoT) signals, radar point clouds, and cognitive workload metrics National Aeronautics and Space Administration Task Load Index (NASA-TLX) to capture a user’s unique interaction signature. A specialised feature-extraction pipeline called unified multi-model behavioural embedding (U-MoBE) converts raw sensor data into statistically and semantically rich representations using spectral, entropy-based, temporal, and spatial descriptors. These features are processed by a custom deep neural network for user behaviour analysis (UserBehaviorNet, UB-NET), which supports both closed-set (unknown) and open-set (registered) user identification via a confidence-based rejection method. The model achieves 100% accuracy on both the training and test sets, with minimal cross-entropy losses of 0.0203 and 0.0167, respectively. A confusion matrix assures perfect classification, while the SHapley Additive exPlanations (SHAP) analysis explains the impact of NASA-TLX components on model decisions. Additionally, inference validation on unknown users exhibits strong generalisation and secure rejection, establishing the framework’s efficacy in real-time authentication scenarios.

Background/Objectives: Heterogeneous imaging protocols, a lack of labeled data, and domain shifts continue to make training deep learning models to analyze medical images a challenge. This study presents ETDACVO (Enhanced Tasmanian Devil Anti-Conservative Variable Optimization), a hybrid evolutionary optimization system designed to improve convergence stability and cross-domain robustness in brain tumor MRI classification. Methods: ETDACVO combines Tasmanian Devil Optimization (TDO), Anti-Conservative Variable Optimization (ACVO), and Exponentially Weighted Moving Average (EWMA) smoothing to stabilize evolutionary parameter updates. Unlike existing approaches that optimize augmentation policies or optimizer dynamics separately, ETDACVO simultaneously evolves both components within a single evolutionary loop. The framework was evaluated on four MRI datasets (Nickparvar, Mendeley, BRISC, and Figshare), comprising 28,151 images. In addition, a convergence-aware explainability mechanism, CA-EA-GradCAM, was developed by integrating gradient saliency, transformer attention, and evolutionary convergence confidence to generate confidence-sensitive tumor localization maps. Results: Experimental results demonstrated that ETDACVO achieved a 2.3–2.5% improvement in classification accuracy and converged 19–22 epochs faster than baseline optimizers. The statistical significance of these improvements was confirmed using paired statistical tests (p < 1 × 10⁻⁵). Cross-dataset transfer experiments further showed strong domain-shift resilience, with performance retention reaching 92.8%. The proposed CA-EA-GradCAM mechanism provided interpretable and confidence-aware tumor localization maps. Conclusions: ETDACVO provides a robust and computationally efficient optimization framework for deep-learning-based medical image analysis. By jointly optimizing augmentation strategies and optimizer dynamics, the framework enhances convergence stability, cross-domain robustness, and interpretability, making it a promising approach for reliable brain tumor MRI classification under heterogeneous imaging conditions.

Dense Industry 5.0 deployments require IIoT task offloading schemes that are simultaneously scalable, low latency, and resilient to wireless eavesdropping under harsh industrial blockage. This paper proposes a unified secure offloading framework that integrates reconfigurable intelligent surfaces (RIS), frequency hopping spread spectrum (FHSS), and uplink nonorthogonal multiple access (NOMA) within a multiple MEC architecture. We formulate a mixed integer nonconvex max min problem that jointly optimizes device server association, computation allocation, transmit power, FHSS carrier selection, and RIS phase shifts to maximize the minimum secrecy rate while satisfying strict end to end latency and minimum rate constraints. To solve the resulting coupled design, we develop a block coordinate descent algorithm with successive convex approximation, yielding tractable subproblems and a monotonic improvement of the objective. Simulations show that the proposed design improves the minimum secrecy rate by 18.75%, reduces average latency by 41%, and improves energy efficiency by up to 30% relative to five benchmarks, including traditional NOMA, random RIS phases, fixed power allocation, no FHSS, and conventional OMA. These results highlight the benefits of jointly exploiting spatial control (RIS), spectral agility (FHSS), and massive access (NOMA) for secure and reliable IIoT offloading in obstructed industrial environments. Furthermore, by ensuring robust connectivity in these harsh settings, this research directly supports the digital transformation of the mining sector, facilitating secure IoT safety monitoring and automated extraction systems in alignment with Saudi Vision 2030.