SRSiwar Rekik

Siwar Rekik

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Siwar Rekik has a total of 16 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    MTGA-Net: A Multi-Tier Global-Attention Deep Learning Framework for Enhanced Breast Mass Classification in Mammographic Images

    by Ala’A R. Al-Shamasneh, Siwar Rekik, Hélène Kanso, Yu Wang

    Breast cancer, recognized as a leading health threat among women globally, underscores the need for innovations in early detection and diagnostic methods. Traditional mammography, despite being widely used, suffers from time-consuming processes and variability in interpretation. This study aims to enhance the accuracy and efficiency of breast mass classification in mammographic images by developing and validating two innovative Computer-Aided Diagnosis (CAD) systems utilizing deep learning techniques. Robust breast mass grading is crucial for diagnosis, reduces unnecessary biopsies, and assists radiologists in making informed clinical decisions during screening and diagnostic workflows. The study introduces the Multi-tier Global-attention Network (MTGA-Net), designed to enhance feature representation by integrating both local feature extraction and global context information. The system is based on the ResNet152V3 backbone and utilizes global pooling combined with multi-layer perceptron operations to extract both local and global features. MTGA-Net includes a branch-attention module that focuses on critical features of the mass, employing a dual-stage transfer learning approach. Initially, the system leverages the ImageNet dataset for general feature extraction, followed by fine-tuning on a specialized local mass patch dataset. The system’s interpretability and reliability are assessed through Gradient-weighted Class Activation Mapping (Grad-CAM) visualizations, offering insight into model decisions. Empirical validation of MTGA-Net on the Digital Database for Screening Mammography (DDSM) and INbreast dataset showed remarkable performance, achieving an AUC score of 0.98375 and 0.99311, respectively. These findings underscore the transformative potential of AI-driven technologies in medical imaging, establishing a new standard for breast cancer diagnosis and patient care.


  • Open Access

    AdaptivePixGuard: Attention-Enhanced Temporal Convolutions for Robust Mobile Pixnapping Detection

    by Siwar Rekik, Sajid Mehmood, Mimouna Alkhonaini

    Pixnapping attacks rely on Android vulnerabilities in the UI rendering pipeline that allows sensitive on-screen pixel data such as OTPs or passwords to be stolen without any screenshot permissions being requested by the attacker. The attacks circumvent conventional security systems, which are very dangerous to mobile users. In this paper, This paper presents AdaptivePixGuard, a framework of real-time pixnapping attack detection and mitigation based on a lightweight yet adaptable AI framework. The framework uses a Temporal Convolutional Network (TCN) with self-attention to take dynamic behavioral sequences such as system calls, overlay events, and pixel entropy changes. AdaptivePixGuard integrates online gradual learning to ensure a desirable detection accuracy in changing threat settings. This Work is tested on CIC-AndMal2017 and CICMalDroid 2020 datasets showing 96.2% for categorization accuracy, AUC of 0.98% and CPU overhead of less than 4% on mid-range Android devices with mitigation executed in 180 ms. These findings demonstrate the usefulness of the framework in real-world practice to offer a valuable countermechanism to zero-permission pixel-exfiltration attacks.


  • Open Access

    A Structured Lifecycle Model for Quantum Software Engineering: Bridging Technical Challenges and Future Directions

    by Hessa Alfraihi, Islam Zada, Mahmoud Ahmed, Sara Shahzad, Siwar Rekik, Abdullah Alshahrani, Nguyen Phi

    Quantum computing is one of the research areas progressing rapidly toward practical deployment, yet the engineering of scalable and reliable quantum software remains underdeveloped. Current quantum software engineering (QSE) practices are largely tools‐driven and ad hoc that providing limited support for managing probabilistic execution, hybrid quantum–classical workflows, noise sensitivity, and hardware constraints. This study proposed a structured QSE lifecycle that integrates quantum‐specific characteristics with disciplined software engineering practices and principles. The proposed lifecycle organizes development into six phases, encompassing quantum requirements engineering, formal modeling, architecture and circuit design, hybrid integration, noise‐aware testing, and deployment with monitoring. Each phase is supported by explicit artifacts and quantitative criteria to enable systematic progression and iterative refinement. The QSE is validated through expert assessment and simulation‐based experimentation using representative variational quantum algorithms under the realistic noise conditions. The results show improved fidelity convergence, reduced resource overhead, enhanced development stability (DS), and more reliable validation compared with unstructured workflows, demonstrating the value of lifecycle‐driven engineering for quantum software systems.

  • Open Access

    A Hybrid Graph Neural Network and Neural ODE Model to Intrusion Detection in Dynamic Network Topologies

    by Siwar Rekik, Sajid Mehmood

    The adaptive nature of evolving cyber-threats, particularly in Internet of Things (IoT) and Software-Defined Networking (SDN) environments, necessitates intrusion detection systems (IDS) that can adjust to highly dynamic network topologies. Traditional machine learning-based IDS models are limited in their ability to adequately capture discrete temporal dynamics and complex structural dependencies, and therefore struggle to accurately detect sophisticated attacks. In this paper, we propose a new intrusion detection system that employs Graph Neural Networks (GNNs) and Neural Ordinary Differential Equations (Neural ODEs) to represent network traffic as a dynamical graph that evolves over time. To enhance efficiency, noise reduction is achieved through a smoothing-sharpening filter, class imbalance is addressed using the Synthetic Minority Oversampling Technique (SMOTE), and feature selection is integrated via the Tyrannosaurus Optimization Algorithm (TOA). Model parameters are further fine-tuned with the Capuchin Search Optimization Algorithm (CSOA) to optimize classification performance. The proposed framework achieves an accuracy of 98.5%, a precision of 98.2%, a recall of 97. 8%, and an F1 score of 98.0%, outperforming state-of-the-art baselines such as CNN, LSTM, and StrucTemp-GNN by up to 5.2% in the experimental analysis of the CIC IoT 2023 dataset, which includes multiple attack scenarios. Statistical testing confirms that these improvements are significant (p < 0.05). There are also extended experimental assessments, involving cross-dataset testing on the ToN-IoT dataset, as well as an ablation, scalability, and topology analysis, and exhibit the strength, generalization, and practicality of the proposed Hybrid Neural ODE-GNN framework. The results indicate that combining topological modeling with GNNs and temporal behavior learning through Neural ODEs provides a highly scalable and dynamic intrusion detection solution for real-time IoT and SDN environments.


  • Open Access

    Steganography in IoT: A Comprehensive Survey on Approaches, Challenges, and Future Directions

    by Maha Driss, Lamia Berriche, Safa Ben Atitallah, Siwar Rekik

    The Internet of Things (IoT) has raised significant security concerns, especially with regard to secure data transfer among resource-constrained devices. While effective, traditional encryption techniques are often computationally expensive and easily identifiable, making them unsuitable for many IoT applications. Steganography is an intriguing approach that allows hiding sensitive information within seemingly ordinary data, preventing unauthorized parties from detecting and accessing it. Existing studies on IoT security lack a comprehensive analysis of steganographic techniques tailored for IoT-specific constraints. This study bridges this gap by providing a comprehensive review of steganographic methods for IoT, exploring approaches across many domains such as spatial, frequency, and hybrid methods used in images, videos, audio, text, network traffic, hybrid approaches, and quantum-based methods. We evaluated and validated the techniques employed using key metrics such as imperceptibility, robustness, and embedding capacity while emphasizing the computational restrictions, real-time processing requirements of IoT devices, and the vital requirement for energy-efficient algorithms. This survey also investigates the integration of steganography and cryptographic approaches, as well as advances in Machine Learning (ML), Deep Learning (DL), and quantum techniques that could revolutionize the field. The study ends with future research directions that underscore the importance of innovative steganographic techniques that strike a compromise between security, efficiency, and scalability in IoT applications.