AI-DRIVEN SENSITIVITY-AWARE HYBRID CRYPTOGRAPHY FOR SECURE MEDICAL IMAGE TRANSMISSION

Abstract

Reliable and efficient medical image transmission is a demanding issue in smart healthcare systems. The existing cryptographic methods employ equal-strength encryption, tending to incur unreasonably high computational overhead for low-risk data or providing deficient protection for highly sensitive data. To overcome this shortfall, we introduce an AI-based sensitivity-aware hybrid cryptographic system capable of adaptively choosing encryption strength depending on medical image sensitivity levels. Lightweight neural classifier, which uses handcrafted statistical features and deep representations, is developed to assess image sensitivity through the analysis of entropy, texture complexity, and structural information. In accordance with the estimated sensitivity, three encryption approaches are utilized: AES-128 for low sensitivity, AES-256 and Elliptic Curve Cryptography (ECC) for medium sensitivity, and AES-256 and Learning With Errors (LWE) for sensitive images.

The model was verified on various medical image modalities such as Diabetic Retinopathy (DR) images, Brain Tumor MRI images, and Dermoscopic skin cancer images datasets. Experimental outcomes indicate that the developed model provides robust performance under various NPCR, UACI, entropy, PSNR, and SSIM measures, maintaining both high security and fidelity of image quality. In contrast to traditional fixed-strength cryptographic techniques, our adaptive method showcases notable gains in computational performance, conserving encryption time with uncompromised confidentiality for sensitive data. These results underscore the promise of sensitivity-aware hybrid cryptography to improve secure medical image transmission, providing a scalable and smart solution for contemporary healthcare networks.