A Privacy-Preserving Optimization of Neighborhood-Based Recommendation for Medical-Aided Diagnosis and Treatment

Nowadays, patients' physiological data and medical records are outsourced to the medical cloud to help the patients and doctors to obtain valuable and reliable services, i.e., diagnosis decision and treatment recommendation, which provides subtle and precise guidance for medical cases. Neighborhood-based recommendation (NBR) provides an effective and vivid approach to finding like-minded nodes to generate recommendations, automating what is usually known as word of mouth. Unfortunately, both diverse medical data and outsource technology raise stringent privacy concerns since the patient data is generally associated with attribute sets which contain somewhat sensitive information, such as age, gender, epidemic, physiological data and clinic result etc. The great privacy-revealing risk will occur if this sensitive medical information is maliciously exploited by network eavesdroppers, untrusted cloud servers or doctors. Considering the above privacy-revealing issues in medical data, this article proposes a privacy-preserving optimization of neighborhood-based recommendation scheme (namely, PPO-NBR), which achieves a secure and privacy-preserving recommendation for medical-aided diagnosis and treatment without revealing the patients' sensitive information. We propose a privacy-preserving medical graph construction protocol and a medical treatment recommendation protocol to implement the secure recommendation on the encrypted medical data, in which it employs BGN Cryptosystem as the building modular to encrypt data and then creatively applies graph theory to expand neighbors to search the best similar features, and uses BLS signature to ensure the authentication, and meanwhile, it deploys a (t, n) oblivious transfer protocol to preserve the confidentiality of recommendation. A couple of experiments are implemented to evaluate the performance in terms of computational costs and communication overheads and indicate that our scheme is efficient and practical for medical-aided diagnosis.

Automated summary

The outsourcing of patients' physiological data and medical records to the medical cloud provides valuable services for diagnosis and treatment, but also raises privacy concerns. This article proposes a privacy-preserving optimization of neighborhood-based recommendation scheme to securely recommend medical-aided diagnosis and treatment without revealing sensitive information. The scheme utilizes encryption, graph theory, BLS signature, and oblivious transfer protocol to ensure security and confidentiality, with efficient performance in computational costs and communication overheads.