NeuroCrypt: Machine Learning Over Encrypted Distributed Neuroimaging Data.

Area of Science:

• Neuroimaging
• Machine Learning
• Data Privacy

Background:

• Neuroimaging data sharing is limited by privacy and regulatory concerns.
• Aggregating distributed datasets improves model generalization but poses technical challenges.
• Existing decentralized methods may lack robust privacy guarantees.

Purpose of the Study:

• To develop a secure and deterministic approach for joint analysis of distributed neuroimaging datasets.
• To enable collaborative machine learning model training without revealing sensitive data.
• To overcome privacy and logistical barriers in multi-institutional data analysis.

Main Methods:

• Utilized secure multiparty computation (SMC) for distributed computation on encrypted data.
• Organizations collaboratively train machine learning models without sharing raw data.
• The approach ensures deterministic computation and does not require a trusted third party.

Main Results:

• Demonstrated the effectiveness of the proposed SMC approach for collaborative model training.
• Empirical evaluations used logistic regression and convolutional neural networks on MRI datasets.
• The method allows joint model training as if data were aggregated, preserving privacy.

Conclusions:

• The proposed secure multiparty computation method facilitates privacy-preserving collaborative machine learning in neuroimaging.
• This approach enhances the potential of distributed datasets for disease detection and biomarker discovery.
• It offers a robust solution for multi-institutional research without compromising data confidentiality.