A lightweight reversible and attack resilient hybrid framework for secure medical image transmission

The increasing use of telemedicine, cloud-based radiology, and electronic health record (EHR) systems has intensified the need for secure and reversible medical image transmission. This paper proposes LDAHF (Lightweight Reversible and Attack-Resilient Hybrid Framework), which integrates transform-domain watermarking, reversible host-image recovery, standardized key derivation, authenticated encryption, and replay protection within a unified architecture. A 16× 16 binary watermark is spatially pre-scrambled using the Arnold Cat Map (ACM) with $$k=5$$ and embedded into the LL₂ sub-band of a two-level Integer Wavelet Transform (IWT) using parity quantization with $$T=20$$ . Three mutually distinct, key-dependent embedding locations are assigned to each watermark bit, and majority voting is employed during extraction. Embedding-induced coefficient residuals are preserved within the protected payload and used to restore the original LL₂ coefficients before inverse IWT reconstruction. HKDF-SHA-256 derives domain-separated encryption and embedding subkeys from a 256-bit master secret, while AES-256-GCM provides authenticated encryption and an authenticated sequence number provides replay protection. Experimental evaluation on representative Brain MRI, Chest X-ray, Head CT, Knee X-ray, and Retinal Fundus images yielded an average PSNR of 31.84 dB and SSIM of 0.8423. Exact host-image reconstruction was achieved for all five evaluated images, with MSErec=0 , Eₘₐₓ=0 , and successful pixel-wise equality. Watermark extraction remained reliable under JPEG compression, Gaussian noise, salt-and-pepper noise, and median filtering, with average NC values of 1.0000, 0.9984, 0.9797, and 1.0000, respectively. In contrast, $$25%$$ cropping and 5^∘ rotation reduced the average NC to 0.4497 and 0.4037, identifying spatial and geometric desynchronization as limitations of the current implementation. Implementation-level tests successfully detected ciphertext tampering, wrong-key authentication attempts, and packet replay across all five experiments. The complete framework achieved an average end-to-end processing time of 91.36 ms per 256× 256 image in a Google Colab CPU-only environment, demonstrating computational feasibility without GPU acceleration. The proposed framework is also aligned with the broader objectives of SDG 3 (Good Health and Well-being), SDG 9 (Industry, Innovation and Infrastructure), and SDG 16 (Peace, Justice and Strong Institutions) through its support for secure digital healthcare, resilient information infrastructure, and trustworthy medical-data exchange.