DeepENC: Deep Learning-Based ROI Selection for Encryption of Medical Images Through Key Generation With Multimodal Information Fusion

With the rapid advancement of the internet and the widespread application of information technology, a large amount of imaging data has been transmitted over the internet in the healthcare domain. The region of interest (ROI) portion of medical-imaging security is important not only for protecting individual privacy but also for accurate clinical diagnosis and treatment. Therefore, an effective security solution is required to prevent third parties from understanding the transmitted data. This paper proposes an efficient image encryption technique called DeepENC that uses multi-modal features to transfer data securely. The first stage performs ROI selection using UNet3+, a deep-learning model with high computational efficiency and fewer network parameters. Subsequently, fingerprint and iris features are extracted, fused and encoded in a deep learning network, and a highly secure encryption key is generated using a novel, 2D hybrid chaotic map. Lastly, the key is employed to encrypt only the ROI portion of the medical images, reducing the time cost. Through comprehensive experimental analysis, this study demonstrates the superiority of the DeepENC technique over other encryption approaches, establishing the validity of the technique for securely transmitting sensitive data.