Protecting Consumer Healthcare Imaging Data from AI-Tampering via Blind Reversible Watermarking

Medical images are increasingly processed across consumer-facing healthcare pathways - teleradiology portals, mobile viewers, MIoT and edge gateways. These images are vulnerable to tampering, privacy abuse, and AI-driven manipulation, such as GAN forgeries and adversarial perturbations, that can mislead diagnosis. We propose a blind, reversible watermarking framework that employs a rotation-invariant biometric fingerprint while preserving diagnostic fidelity through exact reconstruction. The approach generates an Orientation-Guided BSIF (OG-BSIF) signature, protects it with a lightweight FlexenTech permutation, and embeds it in the integer wavelet domain using an adaptive spread-spectrum strategy that combines STDM for robustness with Local Histogram Shifting (LHS) for bit-exact recovery. Experiments on multi-modal datasets, including CT, MRI, ultrasound, and X-ray, achieve high imperceptibility, with average PSNR > 54 dB, SSIM ≈0.997, and perfect reversibility, with a reconstruction PSNR of 90.28 dB. OG-BSIF preserves > 99.9% of features under severe rotation, enabling reliable authentication, and the framework remains robust under FGSM and PGD perturbations, with sub-second execution supporting real-time deployment in consumer and point-of-care settings.