Deep learning based medical image segmentation for encryption with copyright protection through data hiding

Abstract

The prevention of medical information leakage has gained significant attention in recent times. As a result, numerous image encryption schemes are gaining prominence in protecting the privacy of original images. However, third-party users can easily compromise and access encrypted data after decryption. Therefore, it is imperative to develop encryption systems with enhanced confidentiality to address this issue. To tackle these problems, 3D-chaos-based encryption combined with copyright protection is proposed. This achieves high security at a low time cost. The method first segments the most significant information, i.e. the region of interest (ROI) part of the medical image, through the recent deep learning-based segmentation, i.e., you only look once (YOLO) version 8, for image encryption. The 3D-chaos-based encryption encodes only the ROI part, making it well-suited for secure healthcare with a low time cost. Finally, the hash of the ROI and the MAC address of the sender system is embedded into the non-region of interest (NROI) part of the image, making it effective against copyright violation, high bandwidth and storage costs. The results of extensive experiments on COVID-19 and COCO2017 datasets indicate that the method is highly secure, cost-effective and resistant to brute-force attacks. Given the advantages of encryption and data hiding, the proposed method could be an apt choice for medical data transmission and protection against any brute-force, statistical or differential attacks.