A hybrid deep learning framework utilizing CNN, GAN, and autoencoder for safe medical image encryption in IoMT

Abstract In the era of digital healthcare, ensuring the security and privacy of medical images transmitted over Internet of Medical Things (IoMT) networks is paramount. This paper proposes a novel hybrid image encryption framework that integrates Convolutional Neural Networks (CNN), Generative Adversarial Networks (GAN), and Autoencoder to achieve robust, end-to-end medical image protection. Unlike traditional methods, our approach leverages deep feature extraction, adversarial learning, and dimensionality reduction to provide high-security encryption while maintaining reconstruction fidelity. The novelty lies in combining chaotic key generation with deep learning-based encryption and compression, making the system highly resistant to brute-force and statistical attacks. The proposed model introduces a dual GAN structure for encryption and decryption tasks, effectively disguising original images into target domain representations. To evaluate performance, we conduct a comparative analysis with recent state-of-the-art encryption schemes based on metrics such as Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), Entropy, and Correlation Coefficient. Experimental results on public medical image datasets demonstrate that our method outperforms existing techniques, achieving a PSNR of 38.2 dB and SSIM of 0.96 in decrypted image quality, while maintaining high entropy (7.99) and low correlation (