Secure Image Classification Using Deep Learning

Machine learning and security are the buzzwords these days. Just like other fields, privacy concern is a major issue in machine learning systems as well. Current privacy techniques focus on allowing multiple input parties to collaboratively train machine learning models without releasing their private data in its original form. One of the most sensitive data in this regard is medical images. Usage of such data for collectively training models might be against the policies of hospitals, which assure patients that their information would be kept confidential. In such a scenario, privacy preserving machine learning poses several advantages over the conventional methods. In this paper, we have implemented a secure machine learning model based on the multi-party protocol described in SecureML (Mohassel and Zhang in 2017 IEEE symposium on security and privacy. IEEE, pp 19–38, 2017, [1]), on the medical dataset of X-ray images for pneumonia. The performance of these privacy preserving techniques against conventional machine learning algorithms is evaluated.