Adaptive and Lightweight Security Framework for Medical Images Using Intermittent Encryption and Deep Learning in IoT-Enabled Healthcare

Abstract

In this paper, a novel lightweight security framework, integrating a code-based on-off encryption with an authentication based on Convolutional Neural Network (CNN), is proposed to provide security for the medical image transmission and storage in IoT. The system encrypts only the key important portions of the image using BCH codes and LFSR-generated keys, thereby highly improving encryption speed while maintaining image privacy. We test our approach experimentally on the NIH Chest X-ray14 dataset. The decrypted image diagnostic quality PSNR was 41.82 dB and SSIM was 0.986 respectively, suggesting a better resistance against statistical and differential attacks by the encrypted images. The efficiency tests revealed an encryption time of 26.3 ms and a decryption time of 27.5 ms, which makes this work feasible to real-time IoT applications. Additionally, CNN-based integrity verification performed 93.7% correct classification on decrypted images and 98.9% authentication with low false acceptance (1.1%) and false rejection (1.3%). Comparing to AES and chaos-based encryption schemes, the proposed scheme boasts excellent speed, security, and minimum resource overhead, rendering it very suitable for limited-resource healthcare devices.