Fig. 5: Privacy enhancing technologies (PETs) for biomedical data aim to obscure individual identity, sensitive attributes and group memberships from privacy attacks.
From: Addressing contemporary threats in anonymised healthcare data using privacy engineering
PETs have varying strengths and weaknesses and should be tailored to the specific task and data type. a Differential privacy generates processed data which are statistically indistinguishable from the raw population data, but cannot be used to infer details on any individual. b Homomorphic encryption enables analyses on encrypted data without first de-encrypting them and can be used for individual-level analysis. c Secure Multiparty Computation split access between users so that no one has access to the entire dataset, and may be valuable for distributed tasks such as clinical trials. d Federated learning integrates data from multiple sources to develop AI models. e Synthetic data have similar statistical properties to raw data but do not reveal individual properties.
