Systematic review on privacy-preserving machine learning techniques for healthcare data
B. Sasirekha, Chellamuthu Gunavathi · Journal of Cyber Security Technology · 2025
The rapid growth of healthcare data, powered by advances in Electronic Health Records (EHRs), where sensors, medical imaging, and genetic technologies, has generated significant prospects in Machine Learning (ML) to improve disease diagnosis, treatment methodologies, and precision medicine. The sensitive and personally identifiable characteristics of medical data engender significant concerns over privacy, security, and adherence to regulations such as HIPAA and GDPR. To maintain ethical accountability and public trust, ML in healthcare must be executed with resilient Privacy-Preserving algorithms. In response to this critical necessity, Privacy-Preserving Machine Learning (PPML) has evolved as a viable method for safeguarding patient data while preserving model performance. This review paper focus on a systematic analysis of cryptographic, non-cryptographic, and hybrid models of privacy-preserving technologies in healthcare systems, which integrates several ways to enhance privacy and utility. Our review examines advanced methodologies such as Homomorphic Encryption, Secure Multiparty Computation (SMPC), Differential Privacy (DP), and Federated Learning (FL). Each technique is assessed according to its technology foundation, benefits, drawbacks, and relevance to particular clinical situations. Ultimately, our paper addresses the significant challenges of using PPML in real-world applications and proposes future research directions to facilitate the development of dependable, scalable, and privacy-aware medical AI systems.