Medical Image Integrity Protection Through U-Net Based ROI Segmentation and Hybrid Integer Wavelet–Quadtree Embedding

Muna M. Jawad, College of Artificial Intelligence Engineering, University of Technology-Iraq, Iraq, BaghdadFollow
Rasha F. Nadhim, College of Electromechanical Engineering, University of Technology-Iraq, Iraq, BaghdadFollow
Noor A. Yousif, College of Artificial Intelligence Engineering, University of Technology-Iraq, Iraq, BaghdadFollow
Ashwaq T. Hashim, College of Artificial Intelligence Engineering, University of Technology-Iraq, Iraq, BaghdadFollow

ORCID

Muna M. Jawad: https://orcid.org/0000-0001-8561-3866

Ashwaq T. Hashim: https://orcid.org/0000-0002-3282-6419

Article Type

Original Study

Abstract

For decades, secure techniques in medical images have focused on securing the sensitive information through limiting direct access to the images themselves. Nonetheless, these methods tend to cause distortion on the images that can affect the diagnostic value and cause loss of crucial information. There arises the fundamental problem of secure embedding of medical data in such a way that does not affect the diagnostic integrity of the image. To overcome this limitation, a secure yet hidden data-hiding mechanism is introduced that is exploited U-Net-based deep learning for accurate Region of Interest (ROI) segmentation. Therefore U-Net architecture is used for localizing ROI automatically at very precise level, to maintain clinically relevant areas with no unintended side effects. A 4D hyperchaotic keyed hash algorithm is implemented on the extracted ROI in order to obtain a strong hash code, embedded with the electronic patient record (EPR) after that. In order to protect the embedding process, quadtree decomposition splits the Region of Non-Interest (RONI) into adaptive blocks, excluding the ROI from manipulation. Discriminative features are also read over each block to figure out the best embedding places. The final embedding is carried out by using an Integer Wavelet Transform (IWT)-based scheme, inserting secure data into the low frequency coefficients of chosen blocks. However, the framework is highly accurate (Dice = 0.972, IoU = 0.948), produces visually pleasing results (PSNR > 67 dB, SSIM > 0.99), exhibits robust performance against noise, compression, and various attacks, ensures perfect reversibility, and provides secure data embedding. Despite its minor inefficiency in resisting geometrical attacks, the scheme allows for embedding of 1024 bits of information into RONI regions with low usage (<3.02%), and surpasses other methods regarding their reliability and image quality preservation.

Keywords

Medical image, Image integrity, Quadtree decomposition, IWT, Hash code, U-Net, 4D hyperchaotic

How to Cite This Article

Jawad, Muna M.; Nadhim, Rasha F.; Yousif, Noor A.; and Hashim, Ashwaq T. (2026) "Medical Image Integrity Protection Through U-Net Based ROI Segmentation and Hybrid Integer Wavelet–Quadtree Embedding," Journal of Intelligent Informatics, Networking, and Cybersecurity: Vol. 2 : Iss. 1 , Article 4.
Available at: https://doi.org/10.65445/3106-1192.1011