In this work, we propose a CAD (computer-aided diagnosis) system using advanced deep-learning models and computer vision techniques that can improve diagnostic accuracy and reduce transmission risks using the YOLOv7 (You Only Look Once, version 7) object detection architecture. The proposed system is capable of accurate object detection, which provides a bounding box denoting the area in the X-rays that shows some possibility of TB (tuberculosis). The system makes use of CNNs (Convolutional Neural Networks) and YOLO models for the detection of the consolidation of cavitary patterns of the lesions and their detection, respectively. For this study, we experimented on the TBX11K dataset, which is a publicly available dataset. In our experiment, we employed class weights and data augmentation techniques to address the data imbalance present in the dataset. This technique shows a promising improvement in the model’s performance and thus better generalization. In addition, it also shows that the developed model achieved promising results with a mAP (mean average precision) of 0.587, addressing class imbalance and yielding a robust performance for both obsolete pulmonary TB and active TB detection. Thus, our CAD system, rooted in state-of-the-art deep-learning and computer vision methodologies, not only advances diagnostic accuracy but also contributes to the mitigation of TB transmission risks. The substantial improvement in the model’s performance and the ability to handle class imbalance underscore the potential of our approach for real-world TB detection applications.

The Internet of Things (IoT) shares the idea of an autonomous system responsible for transforming physical computational devices into smart ones. Contrarily, storing and operating information and maintaining its confidentiality and security is a concerning issue in the IoT. Throughout the whole operational process, considering transparency in its privacy, data protection, and disaster recovery, it needs state-of-the-art systems and methods to tackle the evolving environment. This research aims to improve the security of IoT devices by investigating the likelihood of network attacks utilizing ordinary device network data and attack network data acquired from similar statistics. To achieve this, IoT devices dedicated to smart healthcare systems were utilized, and botnet attacks were conducted on them for data generation. The collected data were then analyzed using statistical measures, such as the Pearson coefficient and entropy, to extract relevant features. Machine learning algorithms were implemented to categorize normal and attack traffic with data preprocessing techniques to increase accuracy. One of the most popular datasets, known as BoT-IoT, was cross-evaluated with the generated dataset for authentication of the generated dataset. The research provides insight into the architecture of IoT devices, the behavior of normal and attack networks on these devices, and the prospects of machine learning approaches to improve IoT device security. Overall, the study adds to the growing body of knowledge on IoT device security and emphasizes the significance of adopting sophisticated strategies for detecting and mitigating network attacks.

Bug triaging—the process of classifying and assigning software issues to appropriate developers—is a critical yet challenging task in large-scale software development. Manual triaging is time-consuming, inconsistent, and prone to human bias, which often delays issue resolution and misallocates developer resources. This study explores the application of machine learning to automate and improve bug triaging efficiency and accuracy. Using a dataset of over 122,000 issues from the microsoft/vscode GitHub repository, we evaluate several machine learning models including Bidirectional LSTM, CNN-LSTM, Random Forest, and Multinomial Naive Bayes. Our primary contribution is the development of an Augmented Bidirectional LSTM model that integrates enriched textual features and contextual metadata. This model, optimized using Optuna, outperforms traditional baselines, achieving a Micro F1-score of 0.6469 and Hamming Loss of 0.0133 for label prediction, and a Micro F1-score of 0.5974 with Hamming Loss of 0.0062 for assignee recommendation. In addition to demonstrating strong predictive performance, we present a robust end-to-end pipeline for data preprocessing, augmentation, model training, and evaluation using multi-label classification techniques. The study highlights how deep learning architectures, in combination with feature engineering and hyperparameter tuning, can provide scalable and generalizable components to support the automation of bug triaging. These findings contribute to the growing field of intelligent software maintenance by offering data-driven approaches that can support developer workflows and improve issue management efficiency in open-source environments.

The cross-border use of health data is still limited by fragmented data models, weak guarantees around who is requesting access, and inconsistent mechanisms for obtaining and revoking patient consent. The European Health Data Space (EHDS) and the European Digital Identity (EUDI) framework aim to address these issues by giving citizens a portable digital identity and legally valid, auditable control over secondary use of their clinical data, but concrete, national-scale implementations are still missing. We present and evaluate a digital health wallet architecture for Norway that: 1) issues verifiable credentials to both patients and healthcare professionals; 2) records, enforces, and revokes patient consent on a permissioned distributed ledger; 3) verifies access rights using zero-knowledge proofs; and 4) bridges open Electronic Health Record (openEHR) repositories to EHDS-compliant Fast Healthcare Interoperability Resources (FHIR) bundles. The system is built on Hyperledger Besu, uses Groth16 proof circuits with Poseidon hashing for selective disclosure of consent, and encodes General Data Protection Regulation (GDPR) obligations directly into smart contract logic. The architecture was validated in a simulated hospital environment using synthetic ICU-style data. Professional onboarding and role verification completed in approximately 1.2 s (mean, std. 0.2 s). End-to-end consent issuance, verification, and on-ledger recording completed in under 3 s. On-chain verification of zero-knowledge proofs averaged 0.42 s (std. 0.09 s), supporting near real-time authorization without exposing patient identity. The FHIR–openEHR bridge achieved 98.5% mapping success between openEHR compositions and EHDS-mandated FHIR bundles. We further mapped each component to GDPR Articles 5, 7, 25, and 30, demonstrating explicit, revocable, and auditable consent with privacy by design. These results show that an EUDI-compatible digital health wallet can deliver verifiable identity, revocable consent, privacy-preserving auditability, and semantic interoperability fast enough for clinical use. Unlike prior blockchain proposals for consent management, this work demonstrates not only conceptual alignment with EHDS, but an implemented, measured architecture with national deployment potential.