MAMohammed Abdullah Salem Al-Hussaini

Mohammed Abdullah Salem Al-Hussaini

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

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Mohammed Abdullah Salem Al-Hussaini has a total of 22 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Contamination-Aware Cross-Source and Cross-Architecture Benchmarking of Explainable AI for Brain-Tumor MRI Classification

    by Yousuf Nasser Al Husaini, Mohammed Abdulla Al Husaini, Ahmed Gawish, Mohammad Abrar, Khawla Al Hassan, Maya Dawood

    Brain-tumor MRI classification and explainable artificial intelligence (XAI) evaluation can be affected by hidden source overlap and architecture-dependent explanation rankings. This study presents a focused contamination-aware, cross-source, and cross-architecture sensitivity study for brain-tumor MRI classification and post-hoc XAI. Three ImageNet-pretrained classifiers—ResNet-18, DenseNet-121, and ConvNeXt-Tiny—were trained exclusively on 3064 Cheng contrast-enhanced T1-weighted MRI slices using three random seeds. External evaluation used 4793 BRISC tumor images. A preceding image-level audit flagged 1070 BRISC images as possible or probable Cheng-related overlaps; therefore, performance was reported on BRISC-Full, an audit-retained subset of 3723 images, and the 1070-image flagged cohort. DenseNet-121 provided the strongest and most reproducible cross-source classification, achieving 90.34 ± 1.77% accuracy, 90.25 ± 1.79% macro-F1, and 0.9927 ± 0.0009 macro ROC-AUC on BRISC-Conservative. Removing flagged images reduced macro-F1 by 1.05–1.88 percentage points across architectures, while the flagged cohort was systematically easier, showing sensitivity to cohort selection without establishing a causal leakage effect. A comprehensive 15-method XAI benchmark on a pre-specified ResNet-18 showed that no technique dominated faithfulness, stability, and efficiency simultaneously: HiResCAM and LayerCAM offered the strongest overall quality–efficiency balance, whereas ScoreCAM achieved excellent stability at markedly higher computational cost. An eight-method common-correct cross-model benchmark revealed significant rank agreement between ResNet-18 and DenseNet-121 (Spearman ρ = 0.731, p = 0.040) and weaker estimated agreement involving ConvNeXt-Tiny, where Integrated Gradients became the leading method. Additional five-fold patient-separated Cheng evaluation yielded mean macro-F1 values of 0.913–0.936. In a distinct 12-method panel of 300 held-out Cheng images, overall rankings on 298 complete images gave ResNet–DenseNet agreement of 0.965 (95% CI 0.867–0.970), compared with 0.098 and 0.147 for pairs involving ConvNeXt. XAI features did not consistently improve a patient-separated error-referral task. These results support architecture- and protocol-specific interpretation of explanation reliability within the tested datasets.

  • Open Access

    Operational Architectures for Hybrid AI-Based Network Intrusion Detection: A Critical Review and Multi-Axial Taxonomy

    by Alaa Haddad, Sarvesh August Wajs, Mohamed Hadi Habaebi, Mohammed Abdullah Salem Al-Hussaini

    Artificial intelligence is widely used in network intrusion detection, but evidence regarding operational architecture remains fragmented across model families, fusion patterns, deployment environments, and evaluation practices. We conducted a PRISMA 2020-compliant systematic review of secondary AI-NIDS reviews published from 1 January 2020 to 22 September 2026. OpenAlex and arXiv were searched on 22 September 2026; deterministic DOI/title deduplication and prespecified text rules preceded title/abstract screening, and a single reviewer assessed the retrieved full texts. Review-level limitations were appraised across eligibility, search reproducibility, selection/data collection, evidence traceability, and synthesis appropriateness; no aggregate quality score or statistical pooling was used. Of 8519 identified records, 1091 duplicates and 2256 automatically ineligible records were removed, 5172 records were screened, 101 reports were sought, 25 full texts were assessed, and 22 reviews were included. The synthesis supports a five-axis taxonomy spanning observation, AI intelligence, fusion, deployment, and trust/operations, but also shows weak search reproducibility in 12 of 22 included reviews and absent selection transparency in 15 of 22. The findings provide a bounded architectural framework rather than a universal performance ranking. Limitations include open-index coverage, 76 unretrieved reports, single-reviewer assessment, and the use of primary studies only as separate illustrative validation sources. The review was not prospectively registered; a revision-stage protocol was frozen before the new search, and its amendment is reported. A worked five-axis example, a decision-level contrast between sequential and hierarchical integration, appraisal-weighted confidence statements, and proposed evaluation criteria for each open challenge make the framework directly usable for design and assessment.

  • Open Access

    Agentic AI Automation Framework for Computer Network Monitoring and Management

    by Alif Najwan bin Hishamuddin, Mohamed Hadi Habaebi, Md Rafiqul Islam, Mohammed Abdullah Salem Al-Hussaini

    Autonomous network management requires balancing rapid incident resolution with strict operational safety. This paper introduces a production-inspired, end-to-end multi-agent orchestration architecture for network fault diagnosis and resolution. Built upon an event-driven N8N orchestrator and integrated with live routing daemons, the framework couples the iterative Reasoning and Acting (ReAct) paradigm with topological Retrieval-Augmented Generation (RAG). To prevent unintended state disruptions, proposed remediation commands are evaluated using a dynamic Command Risk Score (RC) and governed by a state-aware, two-way interactive Telegram Human-in-the-Loop (HITL) mechanism. Within the scope of six simulated fault scenarios evaluated on a three-router testbed, the multi-agent framework achieved high diagnostic precision and successful network restoration, significantly mitigating diagnostic hallucinations and reducing required human intervention. Furthermore, the system reduced Mean Time to Resolution (MTTR) compared to human engineering baselines. Furthermore, memory decay-based retrieval yields a 32.6% reduction in token consumption and a 34.4% operational cost saving per remediation event. This architecture demonstrates strong feasibility as a controlled proof-of-concept for automated diagnostic and remediation workflows within localized network environments.

  • Open Access

    Patient-Centric Blockchain-Governed EHR Sharing Framework for Secure Digital Radiography Sensor Data and AI-Assisted Chest X-Ray Screening

    by Alaa Haddad, Mohamed Hadi Habaebi, Mohammed Abdullah Salem Al-Hussaini

    Secure exchange of electronic health records (EHRs) requires clinical confidentiality, verifiable provenance, patient-controlled authorization, and practical handling of high-resolution medical imaging payloads generated by IoMT digital radiography sensors. This study presents an end-to-end patient-centric framework linking digital radiography flat-panel detector streams and acquisition metadata to a blockchain-governed EHR management system. Clinical payloads are secured via authenticated symmetric bulk encryption (AES-256-GCM), stored off-chain on the InterPlanetary File System (IPFS), and referenced on-chain through cryptographic record hashes, ECDSA signatures, wrapped-key references, and stateful access control predicates. A smart contract enforces dynamic, patient-controlled view, edit, and prospective revocation rights. The framework incorporates a DenseNet121 deep learning module that computes clinician-reviewable probability vectors for 14 thoracic findings, cryptographically binding each inference output to the source-image hash, model version, execution timestamp, and reviewing clinician. To prevent data leakage, the NIH ChestX-ray dataset was split at the patient level (80/10/10). System evaluation across 50 benchmark trials demonstrates sub-second execution overhead: encryption and decryption of a 100 MB radiograph payload require 0.957 ± 0.032 s and 0.898 ± 0.028 s, respectively, while IPFS upload and download of a 500 MB imaging payload take 11.470 ± 0.450 s and 13.260 ± 0.520 s. Smart contract state-changing operations consume between 29,150 and 168,010 gas units. The proposed architecture establishes a cryptographically verifiable, high-throughput pipeline for digital radiography sensor data management and AI-assisted EHR decision support.

  • Open Access

    AI-Enhanced Anomaly Detection in Water Treatment Plants

    by Ahmad Ihsan Akmal Izram, Mohamed Hadi Habaebi, Mohammed Abdullah Salem Al-Hussaini

    Industrial water treatment plants are increasingly dependent on cyber–physical systems (CPS) and automated control processes for their operational safety and efficiency. However, the embedding of digital control networks exposes these critical infrastructures to sophisticated cyber–physical attacks, including malicious tampering with chemical dosing units and physical actuators. This paper proposes a robust, AI-enhanced anomaly detection framework designed to identify multi-stage malicious activities in water treatment systems using real-world industrial datasets. The proposed system is developed and validated on the Secure Water Treatment (SWaT) dataset, which contains multivariate sensor and actuator time-series data collected from a fully operational physical testbed under both normal operations and targeted cyber–physical attacks. First, high-frequency sensor noise is filtered, and cross-channel measurement reliability is maximized using a Kalman filter-based sensor fusion module. Subsequently, the fused-state vector is analyzed using an unsupervised Isolation Forest algorithm optimized for high-dimensional boundary isolation. To eliminate false negatives caused by stealthy, low-amplitude data injections that bypass purely statistical models, a deterministic, rule-based verification layer derived from physical process control logic is integrated. By integrating a discrete linear Kalman filter with an unsupervised Isolation Forest and deterministic physical rules, the framework effectively suppresses high-frequency sensor noise, achieving a 67.8% reduction in root mean square error (RMSE), while maintaining high detection accuracy across complex industrial attack scenarios. Experimental results demonstrate that the proposed hybrid framework yields superior detection capability, achieving a Precision of ≈95%, a Recall of ≈93%, a scenario-level F1-score of 94.1% (alongside a sample-level F1-score of 21.5%) and an edge inference latency of 0.6 ms, effectively demonstrating its suitability for deployment within simulated real-time industrial edge computing environments. The findings further confirm that combining statistical machine learning, state-space sensor fusion, and invariant physical process logic provides a resilient defense paradigm for securing critical industrial infrastructure against modern cyber–physical threats.


  • Open Access

    Hybrid Watermarking and Adaptive Misinformation for Protection Against AI Model Extraction in Edge-Deployed Cyber-Physical Security

    by Fatimah Azzahrah Binti Razali, Mohamed Hadi Habaebi, Mohammed Abdullah Salem Al-Hussaini

    Artificial intelligence (AI) models are increasingly deployed in edge-deployed cyber-physical security systems for tasks encompassing monitoring, threat classification, and automated decision-making. While these models offer robust performance, their deployment through open or semi-open Machine Learning as a Service (MLaaS) interfaces exposes them to severe security threats, prominently model extraction attacks. In such attacks, an adversary systematically queries a target API to replicate the victim model’s behavior. This study proposes a novel hybrid defense framework combining Adaptive Misinformation (AM) and Trigger-Based Watermarking (WM) to protect AI models against black-box extraction. Utilizing a LeNet architecture, the victim model was trained on the MNIST dataset, while a simulated attack utilized 50,000 EMNIST samples to train a clone model. The framework employs Maximum Softmax Probability (MSP) for out-of-distribution (OOD) detection to identify suspicious queries and strategically inject misleading responses, alongside a fine-tuned embedded watermark for ownership verification. Experimental evaluations using 10-fold cross-validation reveal that the baseline extraction attack yielded a clone model accuracy of 96.32%. Upon implementing the AM + WM framework, clone model accuracy degraded significantly to 53.67%, while the victim model maintained an accuracy of 98.97%. Furthermore, the protected model achieved a 100% Trigger Match Rate (TMR), ensuring reliable intellectual property verification. The proposed framework provides a prototype validation for lightweight edge architectures to balance security, model utility, and ownership protection in cyber-physical deployments.

  • Open Access

    A Multi-Seed Analysis of Adversarial Vulnerability in BiLSTM Continuous Authentication

    by Ahmed Mahfouz, Mohammed Abdulla Salim Al Husaini, Alaa A. K. Ismaeel, Yousuf Al Husaini

    A single user-invariant tensor, kinematically impossible for any human finger to produce, bypasses bidirectional long short-term memory (BiLSTM) continuous-authentication defenders with numerically identical structure across four independently trained generators. We arrive at this finding by training generative adversarial networks against BiLSTM defenders on 51 users across three independent random seeds, with the data partition held fixed, to test the prevailing assumption that successful generative attacks must reproduce the victim’s kinematic behavior. Aggregate attack success rate varies from 31.4% to 45.1% across seeds, a 13.7 percentage-point spread arising purely from optimization stochasticity, demonstrating how unreliable single-seed reporting is as an estimator of the true attack surface. A four-group descriptive stratification shows that 8% of users are attacked across all three seeds, 31% are consistently safe, and 61% exhibit seed-dependent outcomes. Classifier accuracy on zero-effort impostors does not predict adversarial vulnerability (Spearman ρ=−0.058, permutation p=0.688), whereas intra-user behavioral variance does (ρ=+0.351, permutation p=0.012, Bonferroni-corrected). The mechanism is not behavioral emulation but convergence to an Adversarial Skeleton Key, a tensor located in an unregularized region of the BiLSTM’s decision surface that the network reliably maps to acceptance, despite lying many standard deviations outside any genuine human distribution. The mimicry-centric evaluation paradigm underestimates the real threat surface. Input-space plausibility must be treated as a defensive layer rather than a preprocessing concern.

  • Machine Learning and Deep Learning Models for Early Identification of At-Risk Students Using OULAD VLE Data

    by Al Husaini, Yousuf Nasser, Mohammed Abdulla, Al Kishri, Wasin Ayman, Mahmood Al Bahri, Nur Syufiza Ahmad Shukor

  • Open Access

    Automated Classification Radiograph of Periodontal Bone Loss Using Deep Learning

    by Mohammed Abdulla Salim Al Husaini, Mohamed Hadi Habaebi, Seema Yadav

    Background: Periodontitis is a common chronic inflammatory condition of the supporting tissues of the teeth that destroys the tissues and, if left untreated, results in tooth loss. Accurate and early classification of periodontal bone loss through dental radiographs, such as orthopantomograms (OPGs), is crucial for effective diagnosis and treatment planning. Objectives: The present study aimed to evaluate and compare 3 deep learning architectures—InceptionV3, InceptionV4, and ResNet-50—for classifying OPGs into distinct grades of dental features characterised by periodontal bone loss. Design: A comparative experimental design was adopted to analyse the performance of multiple convolutional neural network architectures trained on OPG images representing various grades of periodontal conditions. Methods: A deep convolutional neural network architecture with varying filter and feature layers was implemented. The training process was conducted using MATLAB on a Dell computer equipped with a GeForce RTX 4060 GPU. Image data augmentation was applied to increase dataset diversity. Several combinations of epochs, learning rates, and optimisation algorithms were tested to enhance performance. Model evaluation metrics included accuracy, precision, recall, and F 1-score. Results: Among the tested architectures, ResNet-50 achieved superior performance, reaching an accuracy of 96.8% by the 16th epoch when trained using an SGD optimiser with momentum and a learning rate of 0.001. It also demonstrated higher precision, recall, and F 1 scores compared to InceptionV3 and InceptionV4, confirming its effectiveness in OPG classification. Conclusion: The findings indicate that ResNet-50 provides better classification accuracy and reliability than InceptionV3 and InceptionV4 in detecting periodontal bone loss from OPG images. Expanding the dataset and exploring advanced data augmentation and hyperparameter tuning could further improve model robustness. This study highlights the potential of deep learning-based OPG classification systems to assist dental professionals in faster and more accurate detection of periodontal diseases.


  • Open Access

    A systematic review of the utilization of artificial intelligence applications within the educational Oman systems

    by Yousuf Nasser Said Al Husaini, Hamed Hilal Nasser Al Yahmadi, Mohammed Abdulla Salim Al Husaini

    In the wake of rapid digital transformation and the increasing role of artificial intelligence (AI) in enhancing educational outcomes, this study investigates the integration of AI applications within the Omani educational portal under Oman’s Vision 2040. A systematic review was conducted using the Education Resources Information Center (ERIC) database, focusing on peer-reviewed articles published between 2020 and 2024. The six studies were selected for analysis following a rigorous screening process with defined inclusion and exclusion criteria. The findings indicate that AI is currently enhancing personalized learning, enabling data-driven decisionmaking, and supporting hybrid education platforms in Oman. These developments reflect the educational system’s readiness to adopt further AI technologies while directly contributing to the strategic priorities of Oman’s Vision 2040. This study provides important insights for decision-makers and educators looking to enhance the incorporation of advanced AI applications, thus revolutionizing educational practices and fostering sustainable national growth.