FSSR: Fine-Grained EHRs Sharing via Similarity-Based Recommendation in Cloud-Assisted eHealthcare System

Cheng Huang, Rongxing Lu, Hui Zhu, Jun Shao, Xiaodong LinAsiaCCS 2016 · DOI

TL;DR

This paper introduces FSSR, a novel system for secure and fine-grained sharing of Electronic Health Records (EHRs) in cloud-assisted e-healthcare. It tackles the critical challenge of balancing patient privacy with the need for doctors to access relevant health information by combining privacy-preserving doctor recommendation with attribute-based encryption for fine-grained access control.

What Problem Does This Paper Address?

The increasing digitization of healthcare has led to the widespread adoption of Electronic Health Records (EHRs). While EHRs offer numerous benefits, their sensitive nature raises significant privacy and security concerns. Patients need to be able to share their health data with healthcare providers, but this sharing must be controlled to prevent unauthorized access and potential misuse. Existing solutions often struggle with providing both fine-grained control (allowing patients to specify exactly which parts of their EHRs are shared with whom) and efficient, privacy-preserving mechanisms for finding suitable doctors. Patients need a way to discover appropriate doctors without revealing sensitive personal information, and doctors need to access relevant patient data without compromising privacy.

Methods

FSSR addresses these challenges through a multi-phase approach:

  • System Initialization: The Healthcare Center (HC) sets up the system, generates cryptographic keys, and establishes secure channels. Patients and doctors register, providing their public keys and information. Crucially, HC creates a secure hash index of doctor information and outsources it to a Cloud Server (CS).
  • Privacy-Preserving Doctor Recommendation:
    • PatientQuery: A patient encrypts their demand criteria (attributes) and sends it to the CS.
  • DoctorMatch: The CS uses Locality Sensitive Hashing (LSH) to efficiently find potential doctor matches based on the encrypted demands and the doctor information index. LSH is a technique that hashes similar high-dimensional vectors into the same buckets, enabling fast approximate nearest-neighbor searches. The CS then sends encrypted similarity scores to the HC.
  • DoctorRecommend: The HC decrypts the similarity scores, ranks the doctors, and recommends a list of suitable candidates to the patient. This process ensures that neither the patient's demands nor the doctors' detailed information are directly exposed to the CS.
  • Fine-Grained Privacy-Preserving EHRs Sharing:
    • PatientAuth: The patient selects a doctor from the recommended list and authorizes sharing of specific EHRs based on a defined access policy (attributes).
  • RekeyGen: The HC generates a re-encryption key based on the patient's authorization and the selected doctor's attributes. This key is sent to the CS.
  • EHRReEnc: The CS uses the re-encryption key to transform the patient's encrypted EHRs, making them accessible to the authorized doctor. This step is crucial for fine-grained control, as only EHRs matching the specified attributes are re-encrypted.
  • DoctorDec: The authorized doctor uses their private key to decrypt the re-encrypted EHRs.

The core novelty lies in the combination of LSH for privacy-preserving doctor recommendation and attribute-based encryption (specifically, attribute-based conditional proxy re-encryption) for fine-grained EHR sharing, all orchestrated within a cloud-assisted framework.

Datasets

The paper does not explicitly mention using pre-existing public datasets for training or evaluation. Instead, they use a public XML-formatted EHR file (635KB) from the VA Personal Health Record Sample Data for evaluating the computational cost of patient storage. They also generate synthetic data for doctor information and patient demands for performance evaluations.

Key Insights and Ideas

  • Bridging the Gap: The paper successfully bridges the gap between the need for efficient doctor discovery and the imperative of patient privacy. LSH is a key enabler here, allowing for similarity matching without revealing raw sensitive data.
  • Fine-Grained Control Beyond Roles: FSSR moves beyond traditional role-based access control by allowing patients to define access policies based on EHR content attributes, offering a more granular level of control.
  • Cloud Offloading for Efficiency: Heavy computational tasks are offloaded to the cloud server, making the system efficient for resource-constrained patient and doctor devices.
  • Balancing Efficiency and Accuracy: The use of LSH for doctor recommendation offers high efficiency but comes with a trade-off in accuracy. The paper acknowledges this and suggests linear search on ciphertext as an alternative for higher accuracy when needed, though at a performance cost.

Key Results

The paper presents extensive experimental evaluations demonstrating the effectiveness of FSSR:

  • Computational Cost:
    • System initialization is efficient, averaging around 100 ms.
  • PatientStore (EHR encryption) takes approximately 2 seconds for 100 files without access policies, and around 12 seconds with policies (5 attributes). This is considered acceptable as it's a one-time operation.
  • DoctorMatch (CS part of recommendation) is efficient, with running times around 13 seconds even for 100,000 doctors and 50 attributes.
  • DoctorRecommend (HC part of recommendation) is also efficient, taking less than 3 seconds to recommend 100 doctors.
  • Fine-grained sharing operations (PatientAuth, RekeyGen, EHRReEnc, DoctorDec) are also shown to be efficient, with DoctorDec taking less than 1 second to decrypt 100 shared EHR files.
  • Storage and Communication Cost: The paper provides detailed analyses and tables showing the storage and communication costs, indicating they are manageable and scale reasonably with the number of users and data.

Limitations and Future Work

  • LSH Accuracy Trade-off: The paper acknowledges that LSH, while efficient, may not always find the most similar doctor, presenting a trade-off between efficiency and accuracy. For guaranteed accuracy, a linear search on ciphertext is suggested, but this would impact real-time performance.
  • Emergency Situations: The current scheme might not be optimized for emergency situations where immediate access to all relevant EHRs is critical.
  • Collusion Attacks: The security analysis focuses on individual components, and further investigation into FSSR's security under collusion attacks is suggested.
  • Mobile Experiments: The authors plan to conduct smartphone-based experiments to verify FSSR's effectiveness in mobile e-healthcare scenarios.
  • Complex Scenarios: Exploring more complex scenarios, such as handling emergency situations, is a future research direction.

Summary updated

BibTeX
@inproceedings{huang2016fssr,
  author = {Cheng Huang and Rongxing Lu and Hui Zhu and Jun Shao and Xiaodong Lin},
  title = {{FSSR: Fine-Grained EHRs Sharing via Similarity-Based Recommendation in Cloud-Assisted eHealthcare System}},
  booktitle = {AsiaCCS},
  year = {2016},
  doi = {10.1145/2897845.2897870},
}
Figure