AI is transforming cancer screening, tumor segmentation, and precision treatment planning. But a fundamental tension persists: the more powerful the model, the deeper its reliance on massive, highly sensitive datasets — and medical images are among the most private data imaginable.
A new review, "Privacy-preserving medical image processing based on cryptography primitives: A review of methods," published in Intelligent Oncology, provides a comprehensive survey of cryptography-based frameworks for secure medical image processing.
Highlights:
Threat-oriented taxonomy: Mapping privacy leakage across input data, model parameters, and prediction outputs throughout the machine learning lifecycle
Deep dive into three cryptographic primitives: Data encryption, secure multi-party computation , and homomorphic encryption — with representative schemes, strengths, and limitations
Stage-oriented analytical framework: Examining security characteristics, practical applicability, and computational trade-offs across data protection, model training, model security, and inference
Critical analysis of trade-offs: Computational overhead vs. model accuracy vs. security guarantees — including how aggressive quantization and approximation can undermine diagnostic reliability
Future directions: Flexible integrated privacy frameworks, improved ciphertext packing, lightweight encrypted network architectures, and the promise of blockchain and zero-knowledge proofs
A key takeaway: no single technique can fully meet the demands of real-world healthcare systems. Balancing security, efficiency, and clinical usability is the central challenge for deploying privacy-preserving medical AI at scale.
Full article available on ScienceDirect:
https://www.sciencedirect.com/science/article/pii/S2950261626000464
Contact Information for Intelligent Oncology:
LinkedIn: @IntelligentOncology
Bulesky: @Intelligent Oncology
X: @IntelligentOnco
Facebook: @intelligentoncology
Email Address: editorialoffice@intelligent-oncology.net
Official Website: https://www.sciencedirect.com/journal/intelligent-oncology
Submission Link: https://www2.cloud.editorialmanager.com/intonc/default2.aspx
