A novel multi-layer security framework for medical image transmission in cloud-based wireless body area networks (BAN) using XAI-guided obfuscation and WDTW encryption.

The existing frameworks for medical image encryption in the context of BANs have three major limitations: encrypting the whole image consumes enormous amount of processing time; failure to integrate explainable AI for identifying the importance of regions and prioritizing the same for permutation and XORing and inability of verifying the lossless integrity of the decryption process. To tackle these limitations, this paper introduces a novel scheme called weighted dynamic transform encryption (WDTW) for adaptive encryption of medical images. Firstly, a model based on convolutional neural network (CNN) is trained using the medical image data set and used for understanding image features. Once the image features are learned using the model, Grad-CAM heat maps are produced, which are used to generate region importance maps for selectively performing block-wise permutation and XOR operations. Further, to guarantee the lossless integrity of the process, a verification module is introduced based on weighted dynamic time warping. The experimental results indicate that our framework gives NPCR of 99.61%, Entropy of 7.99, and reduces processing latency by 35% when compared to the conventional full image AES encryption method. The encrypted outputs successfully passed all seven NIST SP 800-22 tests and also demonstrated excellent resistance to known plaintext attacks. The minimum entropy, maximum correlation, minimum NPCR and average UACI values were found to be 7.999956, 0.000997, 99.6036% and 33.4681% respectively.