Journal of Cyber Security, Privacy Issues and Challenges

WomenCare AI: Comparative Analysis and Optimization of Image Processing Techniques for Ovarian Condition Classification

2026-06-11T08:08:30+00:00

Ovarian health assessment remains a significant challenge in modern medical diagnostics, where early detection directly influences treatment success and patient outcomes. Medical imaging plays a vital role in identifying ovarian abnormalities; however, variations in image quality, noise presence, and low contrast often hinder accurate interpretation. Within the paradigm of intelligent healthcare systems, the WomenCare AI system is conceptualized as a deep learning-driven framework that enables automated classification of ovarian conditions through integrated image processing and neural network analysis. Traditional diagnostic approaches depend on manual evaluation, which introduces delays and subjective inconsistencies in classification outcomes. The proposed framework utilizes OpenCV-based preprocessing methods, including noise reduction, normalization, contrast enhancement, and image resizing to improve data consistency and structural integrity of ultrasound images. It further incorporates ResNet50 and EfficientNet models to extract deep features and perform comparative analysis based on accuracy, efficiency, and reliability. The dataset comprises 11,784 ovarian ultrasound images, including 6,784 infected and 5,000 non-infected cases sourced from a publicly available FigShare repository. Experimental evaluation demonstrates that the proposed WomenCare AI system achieves the highest classification performance with an accuracy of 95.31%, precision of 93.06%, recall of 94.37%, and F1 score of 93.71%, outperforming baseline models including ResNet50, VGG19, DenseNet121, MobileNetV2, and Modified Vision Transformer. The integration of Grad-CAM visualization further enhances model interpretability by highlighting clinically relevant regions in ultrasound images. The proposed approach contributes toward improving clinical decision support systems by delivering faster, consistent, and scalable diagnostic assistance while advancing research in AI-based healthcare solutions.

An Explainable Deep Learning Framework for Intelligent Medical Diagnosis and Transparent Clinical Decision Support

2026-06-01T07:20:42+00:00

Artificial intelligence (AI) has become an essential technology in modern healthcare, particularly in medical image analysis and clinical decision support systems. Despite the remarkable diagnostic performance of deep learning models, their black-box nature limits clinical adoption due to the lack of interpretability and transparency. To address this challenge, this study proposes an explainable deep learning framework (XDLF) for intelligent medical diagnosis by integrating a modified ResNet-50 architecture with explainable artificial intelligence (XAI) techniques, including gradient-weighted class activation mapping (Grad-CAM) and SHapley Additive exPlanations (SHAP). The proposed framework was evaluated using publicly available medical imaging datasets comprising chest X-ray, histopathology, and COVID-19 radiography images. Comprehensive preprocessing techniques, including normalization, noise reduction, contrast enhancements, and data augmentation, were employed to improve model robustness and generalization capability. Experimental results demonstrate that the proposed framework achieved superior diagnostic performance with 96.8% accuracy, 95.9% precision, 96.3% recall, and 96.1% F1-score, outperforming conventional CNN-based models. Furthermore, the explainability module successfully identified clinically relevant regions and quantified feature contributions, thereby improving interpretability and physician trust in AI-assisted diagnosis. The proposed XDLF provides a transparent, reliable, and scalable clinical decision support framework suitable for future real-world healthcare applications.

Reimagining Cybersecurity Controls: BDSLCCI vs. ISO/IEC 27001, NIST CSF, and COBIT 2019

2026-06-06T05:38:57+00:00

Cybersecurity has become a major concern for organizations due to the rapid growth of digital technologies and the increasing number of cyber threats such as data breaches, ransomware, and phishing attacks. Various cybersecurity frameworks, including ISO/IEC 27001, the NIST Cybersecurity Framework (CSF), and COBIT 2019, provide structured guidelines for managing risks, improving governance, and ensuring regulatory compliance. However, these frameworks are often complex, costly, and resource-intensive, making them difficult for small and medium enterprises (SMEs) to implement effectively. SMEs typically face limitations in terms of budget, expertise, and infrastructure, which increases their vulnerability to cyberattacks. This research paper presents a comparative study of a newly proposed framework, Business Domain Specific Least Cybersecurity Controls Implementation (BDSLCCI), with established frameworks such as ISO/IEC 27001, NIST CSF, and COBIT 2019. The comparison is conducted based on structure, control objectives, implementation methodology, governance integration, risk management, compliance support, and maturity assessment. The study highlights that BDSLCCI provides a domain-specific, cost-effective, and stepwise implementation approach tailored to SME needs. The findings suggest that while traditional frameworks remain essential for large enterprises and regulatory requirements, BDSLCCI offers a practical and scalable solution for improving cybersecurity readiness among SMEs.

Advanced Cyber Incident Prediction Using Attention-based Temporal Fusion Transformer with Dynamic Optimization

2026-05-26T09:39:36+00:00

Cyberattacks are becoming more sophisticated and frequent, creating major challenges for modern digital infrastructures. Conventional cybersecurity systems mainly depend on reactive defense strategies and static detection techniques, which are often inadequate for identifying emerging and complex threats. To overcome these limitations, this research introduces an optimized temporal fusion transformer (OTFT) framework for the classification and prediction of significant cyber incidents (SCI). The proposed approach combines temporal feature extraction, gated residual learning, recurrent processing, and multi-head attention mechanisms to analyze complex cybersecurity event sequences effectively. Unlike traditional sequential models, the proposed framework can learn both immediate behavioral changes and long-term temporal relationships from multivariate cyber datasets. An adaptive dynamic population control-based polar bear optimization algorithm (DPCPBOA) is further integrated to optimize critical hyperparameters, improve convergence speed, and minimize overfitting during training. The framework was evaluated using benchmark cybersecurity datasets containing large-scale attack records and temporal event information. Experimental analysis demonstrates that the OTFT model achieves superior predictive performance compared with conventional machine learning and deep learning approaches, including LSTM and Seq2Seq architectures. The proposed framework achieved high accuracy, precision, recall, and F1-score while maintaining strong robustness and generalization capability. The overall findings indicate that the proposed model can serve as an intelligent and scalable solution for proactive cyber threat forecasting and advanced cybersecurity management.