Researchers from the Technical University of Munich (TUM), Imperial College London, and a non-profit firm called OpenMined released a paper called “End-to-end privacy-preserving deep learning on multi-institutional medical imaging.”
The research showcased PriMIA- Privacy-Preserving Medical Image Analysis that uses encryption for the data obtained from medical imaging. The paper suggests that the new technology is completely free and has an open-source software framework. The team conducted the experiment on pediatric chest X-Rays and used a deep neural network to categorize them.
Medical data safety from cyberattacks
Although there are other widely-used techniques to safeguard medical data, they aren’t foolproof and can be hacked easily. For instance, centralized data sharing methods have failed to safeguard sensitive medical information from cyber attacks.
The new technology takes care of the data by leveraging federated learning, where only the deep learning algorithm is passed on while sharing the health data and not the whole content.
The team also applied secured aggregation to prevent external entities from knowing the source of the algorithm’s training. This method will prevent anyone from identifying the institution where it originated from, keeping the data safe.
Compatible with several formats
To ensure that statistical correlations are coming from the data records and not from individuals contributing data, the team also used another method. As per the paper, the system can work with a wide range of medical imaging data formats. Besides, it is said to be easily configurable and brings along functional upgrades to FL training.
“PriMIA’s SMPC protocol guarantees the cryptographic security of both the model and the data in the inference phase,” states the report.
“Our methods have been applied in other studies, but we are yet to see large-scale studies using real clinical data. Through the targeted development of technologies and the cooperation between specialists in informatics and radiology, we have successfully trained models that deliver precise results while meeting high standards of data protection and privacy,” said Daniel Rueckert, co-author of the study.
