Securing the Classification of COVID-19 in Chest X-ray Images: A Privacy-Preserving Deep Learning Approach
This addresses privacy concerns for patients in medical imaging, but it is incremental as it applies existing encryption methods to a specific domain.
The paper tackles the problem of preserving privacy in deep learning for healthcare by proposing a privacy-preserving approach using partially homomorphic encryption on chest X-ray images, achieving 94.2% accuracy on plain data and 93.3% on encrypted data.
Deep learning (DL) is being increasingly utilized in healthcare-related fields due to its outstanding efficiency. However, we have to keep the individual health data used by DL models private and secure. Protecting data and preserving the privacy of individuals has become an increasingly prevalent issue. The gap between the DL and privacy communities must be bridged. In this paper, we propose privacy-preserving deep learning (PPDL)-based approach to secure the classification of Chest X-ray images. This study aims to use Chest X-ray images to their fullest potential without compromising the privacy of the data that it contains. The proposed approach is based on two steps: encrypting the dataset using partially homomorphic encryption and training/testing the DL algorithm over the encrypted images. Experimental results on the COVID-19 Radiography database show that the MobileNetV2 model achieves an accuracy of 94.2% over the plain data and 93.3% over the encrypted data.