TL;DR
This study aims to guide researchers and practitioners in designing secure, efficient, and privacy-preserving AI systems for healthcare by proposing a structured classification of privacy-preserving methods into four main categories: cryptographic approaches, decentralized learning methods, perturbation-based techniques, and hybrid models.
This paper examines how privacy-enhancing technologies such as synthetic data, federated learning and ‘Secure Data Environments’ can be integrated into artificial intelligence ( AI ) development processes to uphold key data protection principles in the UK GDPR , like storage limitation, data minimisation, purpose limitation, security, and fairness. The analysis highlights how privacy-enhancing technologies offer benefits beyond anonymisation by embedding privacy-by-design values to support responsible innovation and protect sensitive patient data throughout the design, training, and validation of medical AI systems. The paper uses the 2015 DeepMind and Royal Free case as a practical study to realise the practical and legal benefits of privacy-enhancing technologies in medical AI development, particularly involving public-private collaborations. While grounded in the UK context, the findings have broader relevance to the European Union and other international jurisdictions grappling with tensions between data protection and AI development in the healthcare context.
Initial evaluations using various machine-learning algorithms on pre-and post-generalized datasets demonstrate the privacy framework’s effectiveness in mitigating privacy risks while preserving data usability.
Background Biomedical artificial intelligence (AI) requires the integration of privacy-enhancing technologies (PETs) to safeguard sensitive clinical, imaging, and genomic data while preserving analytical utility. Objectives This review critically and systematically maps applications of PETs across the biomedical AI lifecycle in accordance with PRISMA-ScR guidelines and evaluates their technical trade-offs, deployment feasibility, and residual risks. Methods We systematically searched PubMed, IEEE Xplore, ACM Digital Library, and Scopus for studies published between 2015 and 2025. Eligible studies addressed differential privacy, federated learning, secure multiparty computation, homomorphic encryption, or hybrid approaches in biomedical AI. Data were charted on PET type, modality, lifecycle stage, utility metrics, privacy parameters, and deployment considerations. A critical appraisal rubric assessed threat-model adequacy, methodological clarity, reproducibility, privacy–utility transparency, and deployment realism. Additionally, we hand-searched major venues (USENIX Security, NeurIPS, AAAI) and screened Google Scholar for grey literature, applying de-duplication across sources. Results We identified 87 studies spanning clinical decision support, genomics, and medical imaging. From 25,761 initial records, 3,754 underwent title/abstract screening and 1,968 underwent full-text assessment. PETs demonstrated distinct strengths and limitations: differential privacy provided provable guarantees but reduced performance on imbalanced data; federated learning improved data access but remained vulnerable to gradient leakage; and cryptographic methods ensured confidentiality at high computational cost. Synthetic data generation supported privacy-conscious data sharing and benchmarking but remained sensitive to disclosure risk, fidelity loss, and subgroup representation. Hybrid and emerging approaches, including trusted execution environments, zero-knowledge proofs, and privacy-preserving transformer architectures, mitigated composability gaps yet lacked full end-to-end assurance. Case studies at hospital and biobank scale illustrated practical feasibility and infrastructure demands. Conclusions Situating PETs within technical and operational contexts clarifies their capabilities, limitations, and deployment challenges. Residual risks persist, including fairness concerns, inference-time leakage, and overreliance on PETs as compliance proxies. Sustained technical innovation and institutional governance remain essential for the trustworthy integration of PETs in biomedical AI.
The rapid adoption of interactive mobile healthcare applications and cloud-enabled medical services has significantly increased the volume of sensitive patient information stored and exchanged across distributed cloud environments. Mobile health (mHealth) systems frequently rely on cloud-based repositories to support real-time accessibility, remote monitoring, and intelligent healthcare analytics. However, the involvement of third-party platforms and remote data processing introduces serious concerns related to privacy preservation and unauthorized disclosure of medical records. To address these challenges, this study proposes a bio-inspired optimization framework based on the Discrete Grey Wolf Optimizer (DGWO) for effective optimized key generation in healthcare data sanitization processes. The proposed approach integrates perturbation-based privacy preservation with optimized key management to enhance secure data sharing in mobile cloud healthcare environments while maintaining data utility for analytical tasks. The framework is evaluated using four benchmark healthcare datasets under multiple performance measures, including privacy preservation rate, information entropy, resistance against inference attacks, and data utility retention. Experimental results demonstrate that the proposed DGWO-based sanitization mechanism outperforms existing perturbation approaches in achieving a balanced tradeoff between privacy and utility. The proposed model is particularly suitable for interactive mobile healthcare systems that require secure, efficient, and privacy-aware cloud data management.
Sepsis is a life-threatening medical emergency which needs immediate treatment because its early diagnosis in intensive care units (ICUs) proves vital for reducing death rates. The study introduces a sepsis prediction system which protects user privacy through its use of personalized federated learning and differential privacy technology (pFL-DP) within a 5G-enabled intelligent healthcare system that enables both clinical AI growth and environmental preservation. The framework enables hospitals to develop shared models by training their systems together without needing to disclose any actual patient information which protects their data while maintaining complete privacy security. The hybrid feature selection which combines filter-based statistical techniques with wrapper-based optimization methods is implemented to improve the efficiency and interpretability and sustainability of computational resources for MIMIC-IV electronic health record data analysis. Three models: XGBoost, Bidirectional Long Short-Term Memory (Bi-LSTM), and a stacked ensemble—are evaluated under centralized and federated settings. Results show that the personalized federated ensemble achieves 90.1% accuracy, 87% sensitivity, and an average early warning lead time of 5.1 h prior to clinical diagnosis. Performance closely matches centralized learning while preserving patient privacy. Deployment simulations in a 5G-enabled edge–cloud environment demonstrate millisecond-level inference latency and reduced communication overhead, confirming the feasibility of secure, real-time, and sustainable AI-driven sepsis prediction in critical care environments.