Securing medical image privacy in cloud using deep learning network

Abstract

The worldwide usage of Internet of Things (IoT) applications enhanced the utilization of consumer devices, such as smartphones, computers, screening equipment used in hospitals that merely rely on imaging techniques. Numerous images got generated over the cloud platform in a daily basis ad create storage complexity. On the other hand, securing the data stored in the cloud is important. Instead of storing large amount of data into the cloud, lightweight dynamic processing of data suppresses the complex issues in cloud security. Here secure cloud-based image processing architecture is discussed. Privacy preserving medical data communication is considered as the specific research scope. Cryptographic technique used to encode the original data and decode the data at the other end is currently in usage as conventional design. Providing privacy to the medical records through adding noise and denoising the same records is the proposed idea. The proposed work is keenly focused on creating a light weight cloud architecture that communicates the medical data effectively with privacy perseverance using deep learning technique. In the proposed system, the design of an efficient image denoising scheme with a hybrid classification model is created to ensure reliable and secure communication. Deep learning algorithms merged to form a Pseudo-Predictive Deep Denoising Network (PPDD). The proposed system's benefit is ensuring added security in Dark Cloud using a newly structured algorithm. The original data is packed in the Deep cloud using the Gaussian noise act as a key. The complete packing and unpacking of medical data is encapsulated by the transformed images. Over the cloud premise, the data is highly secured and invisible to the malicious users. To reduce the storage complexity, the dynamic data is unpacked and denoise process is applied at the edge devices. During the authorized access period alone, the data is decrypted and accessible at the edge nodes. The maximum process is dynamically happen in the cloud without depending on the storage boundary. The performance of proposed PPDD network model is evaluated through Signal to noise ratio (SNR), Similarity index (SI),Error Rate(ER) and Contrast to noise ratio(CNR). The proposed architecture is comparatively validated with existing state-of-art approach.