Healthcare analytics based on privacy preservation involves data that is stored safely, has controlled access and is trained jointly without revealing sensitive medical records. This paper presents a converged framework, which unites ciphertext-policy attribute-based encryption (CP-ABE), permissioned blockchain storage, and federated deep learning to predict illnesses. Though the underlying datasets (Pima Diabetes and UCI Heart Disease) is public, we emulate an environment reminiscent of a real-life multi-institutional environment, whereby data is logically scattered across multiple healthcare customers. The records of each institution are locally encrypted with CP-ABE and stored only as ciphertext hashes and off-chain pointers on a permissioned blockchain, whereby the confidentiality and integrity are guaranteed and the raw data is not concentrated. It is then trained in a federated MBiLSTM-GRU model across these distributed clients allowing them to share the intelligence without aggregating raw data. The suggested framework will combine secure access control, tamper evidenced storage and privacy preserving collaborative learning. The competitive classification performance on both datasets can be proved through the experiment and the viability of federated learning integration with blockchain-based secure storage in a healthcare setting is also shown.
Patil et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: