Mohamed H. Saad

Automated osteoporosis assessment from knee radiographs remains a challenging computer-aided diagnosis task due to subtle structural differences among normal, osteopenic, and osteoporotic bone patterns, together with the limited availability of transparent and reproducible evaluation frameworks based on publicly accessible datasets. Existing studies predominantly rely on standalone deep learning models or conventional ensemble methods that assume fixed classifier reliability and rarely incorporate predictive uncertainty into decision-level fusion. To address this limitation, this study proposes an Entropy-Guided Adaptive Confidence-Weighted Complementary Deep Ensemble (E-ACW Ensemble) for uncertainty-aware osteoporosis-related knee radiograph classification in both binary and multi-class settings. The proposed framework integrates standardized radiographic preprocessing, fold-specific data augmentation, complementary deep feature learning, and an entropy-guided adaptive fusion strategy. Predictive uncertainty is estimated using Shannon entropy to enable sample-specific confidence weighting during decision-level fusion. The framework was evaluated on publicly available knee radiograph datasets using a rigorous five-fold stratified cross-validation protocol. Comprehensive experimental evaluation, including comparative benchmarking, ablation studies, fold-wise stability analysis, statistical significance testing, and computational efficiency assessment, was conducted to validate predictive performance, robustness, and reproducibility. Among the investigated configurations, the proposed EfficientNetB0 + MobileNetV2 + E-ACW Ensemble achieved the best overall performance, reaching 99.2% accuracy and 99.1% F1-score for binary classification and 98.7% accuracy and 98.5% F1-score for multi-class classification while requiring substantially lower computational cost than the corresponding high-capacity ensemble. These findings demonstrate that entropy-guided adaptive confidence weighting provides an effective and computationally efficient decision-level fusion strategy for osteoporosis-related knee radiograph classification. The proposed E-ACW Ensemble establishes a reproducible framework that can be readily extended to future computer-aided musculoskeletal image analysis and uncertainty-aware medical imaging applications.

The security of critical infrastructure, particularly nuclear facilities, is paramount for public safety. Conventional Closed-Circuit Television (CCTV) surveillance relies on static camera placement, which fails to adapt to dynamic adversarial behavior, resulting in coverage gaps and inefficiencies. This study proposes a novel Adversarial Path Planning (APP) framework that integrates game theory, probabilistic risk assessment, and bilevel optimization to enhance surveillance coverage, intrusion detection, and resource allocation. APP simulates adversarial movement and iteratively refines camera placement to dynamically counter evolving threats, minimizing blind spots and optimizing detection probability. The framework models the facility as a weighted surveillance graph, identifying high-risk intrusion paths to optimize camera positioning for maximum coverage with minimal redundancy. A case study on a hypothetical nuclear power plant demonstrates APP’s effectiveness: it achieves 95% surveillance coverage, improves detection accuracy to 98%, and reduces dead zones by 85%, significantly outperforming conventional methods like the Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Ant Colony Optimization (ACO). Furthermore, APP reduces the required number of cameras by 40% while improving cost efficiency by 27%, underscoring its potential for resource-conscious security optimization. The findings establish APP as a scalable and computationally efficient solution adaptable to nuclear facilities, border surveillance, and other high-risk critical infrastructure. Future research should explore AI-driven real-time threat detection and autonomous surveillance systems to further enhance responsiveness in dynamic security environments.

The security of critical infrastructure, particularly nuclear facilities, is paramount in mitigating potential threats and ensuring public safety. Conventional CCTV surveillance deployment relies on static placement strategies that fail to account for dynamic adversarial behavior, leading to coverage gaps and surveillance inefficiencies. This study proposes a novel Adversarial Path Planning (APP) framework, which integrates game-theoretic modeling, probabilistic risk assessment, and bilevel optimization to enhance surveillance coverage, intrusion detection, and resource allocation. By simulating adversarial movement patterns and iteratively refining camera placement, APP dynamically adjusts surveillance strategies to counter evolving threats while minimizing blind spots and optimizing detection probability. The APP framework models the facility as a weighted surveillance graph, identifying high-risk intrusion paths and optimizing camera positioning to maximize coverage while minimizing redundancy. A case study conducted on a hypothetical nuclear power plant demonstrates APP’s effectiveness in enhancing security resilience, achieving 95% surveillance coverage, improving detection accuracy to 98%, and reducing dead zones by 85%—significantly outperforming conventional methods such as Genetic Algorithm (GA), Particle Swarm Optimization (PSO), and Ant Colony Optimization (ACO). Additionally, APP reduces the required number of cameras by 40% while improving cost efficiency by 27%, highlighting its potential for resource-conscious security optimization. The findings establish APP as a scalable and computationally efficient surveillance optimization solution, adaptable to nuclear security, border surveillance, and high-risk critical infrastructure protection. Future research should explore AI-driven real-time threat detection, autonomous security drones, and deep reinforcement learning-based surveillance adaptation to further enhance threat responsiveness and situational awareness in evolving security environments.

Non-orthogonal multiple access (NOMA) is an emerging promising technology for beyond 5G and 6G networks. However, it is essential to investigate its performance over various fading channels. In this work, we shed light on the efficiency of downlink transmission of NOMA systems over the recently proposed Fluctuating Nakagami-m fading channels whose applications cover vehicle-to-vehicle (V2V) communications, and device-to-device (D2D) communications among others. Specifically, we consider an analytical study on the performance of a NOMA transmission in terms of the outage probability, diversity gain, coding gain, and ergodic capacity. The results show that the diversity gain of each user depends on the total number of multipath clusters ($m$) of the Fluctuating Nakagami-m fading channel, while the coding gain depends on several parameters including $m$, the fading severity parameter ($ms$), power allocation scheme, imperfection of the successive interference cancellation (SIC) scheme, and the QoS constraint. Two power allocation schemes have been investigated and compared. Representative simulations have validated the accuracy of the analytical derivations.

Intelligent transportation systems (ITSs) significantly enhance traffic safety and management globally. A critical component of these systems is vehicle classification (VC), which supports vital applications such as congestion control, traffic monitoring, accident avoidance, etc. Traditional classification algorithms rely heavily on visual or sensor-based data (e.g., radar or image signals), often compromised by adverse weather, poor lighting, or occlusion. To address these limitations, this paper introduces a novel VC technique that leverages seismic data to detect vehicle-generated vibrations, thereby reducing susceptibility to environmental conditions and privacy concerns. We propose a self-supervised contrastive learning approach for seismic signal classification, eliminating the need for labeled data for feature extraction and representation. Our method employs specialized data augmentation techniques to create positive and negative pairs, enhancing feature representation. The encoder network extracts meaningful features from seismic signals while the projection head refines latent space representation. Training with contrastive loss ensures that positive pairs are closely aligned and negative pairs are distinctly separated in the latent space. Experimental results validate the efficacy of our approach, achieving state-of-the-art performance using seismic signal classification tasks with limited training data. Our approach achieves an impressive accuracy of 99.8%, underscoring its potential for robust and precise VC in ITSs using seismic data, particularly in data-scarce scenarios. The code is publicly available at https://github.com/MohamedHassanSaad/Vehicle-Classification.git.

Effective fault diagnosis is critical for the safe and efficient operation of rotating machinery in nuclear facilities. This paper proposes a deep learning-based approach that integrates multi-domain signal analysis and transfer learning to classify rotor conditions as either healthy or faulty. Vibration signals are transformed into 2D images and processed using pretrained models: ResNet50, GoogleNet, and a custom Deep Convolutional Neural Network (DCNN). Signal transforms, including the Fast Fourier Transform (FFT), Fractional Fourier Transform, Short-Time Fourier Transform (STFT), Continuous Wavelet Transform (CWT), and Fractional Wavelet Synchrosqueezing Transform, are applied to enhance feature representation. ResNet50 achieved up to 100% accuracy on the primary dataset and over 99% on the secondary dataset. GoogleNet and DCNN also demonstrated excellent performance, achieving accuracies of up to 100% in specific domains. Additionally, transfer learning using YamNet enabled effective sound-based classification of vibration signals. These results show that using advanced signal processing together with deep learning can lead to very accurate and quick fault detection in important safety situations.