Deep Learning-Based Data Hiding Technique for Medical Images

SAIDI, Hadjer (2026) Deep Learning-Based Data Hiding Technique for Medical Images. Doctoral thesis, Université Mohamed Khider - BISKRA.

Text These_Saidi_Hadjer .pdf Download (7MB)

Abstract

This thesis introduces novel deep learning-based frameworks for secure medical image steganography, addressing the critical challenge of protecting sensitive patient information while preserving diagnostic quality. Medical images present unique security requirements due to their dual nature: containing both critical visual data for diagnosis and sensitive patient metadata, which conventional steganographic techniques do not adequately address. Our work presents two complementary approaches to this challenge. First, we develop a Mask-RCNN based framework that intelligently identifies diagnostically insignificant regions within medical images for strategic data embedding. By combining this regionaware detection with Discrete Cosine Transform (DCT) embedding in the frequency domain, our method achieves remarkable imperceptibility with Peak Signal-to-Noise Ratio (PSNR) values exceeding 115 dB while maintaining high payload capacity. Second, we propose a clinical quality-aware convolutional neural network architecture that leverages an encoder-decoder framework for end-to-end steganography. This approach employs parallel processing paths with scaled residual learning to embed secret medical images within cover images while preserving diagnostic features. Extensive experimentation across multiple medical imaging modalities (CT, MRI) from datasets including MIDRC-RICORD-1B, and IQ-OTH/NCCD, demonstrates that our method achieves a good trade-off between payload and imperceptibility. The frameworks developed in this thesis achieve an optimal balance between embedding capacity, imperceptibility, and robust secret recovery, providing healthcare institutions with effective tools for safeguarding patient privacy while maintaining the integrity of medical diagnostics. This work significantly advances the field of medical image security and establishes a foundation for future innovations in secure healthcare information systems.

Actions (login required)

View Item