Related Experiment Video
Updated: Sep 19, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
ViTU-net: A hybrid deep learning model with patch-based LSB approach for medical image watermarking and
V Nanammal1, S Rajalakshmi2, V Remya2
1Department of Electronics and Communication Engineering, Jeppiaar Engineering College, Chennai, India.
Computers in Biology and Medicine
|June 3, 2025
Summary
ViTU-Net enhances medical image security for pneumonia diagnosis using a Vision Transformer and U-Net architecture. This novel approach ensures data integrity and confidentiality against breaches, improving telemedicine and AI diagnostics.
Area of Science:
- Medical Imaging
- Cybersecurity
- Artificial Intelligence
Background:
- Modern healthcare relies on secure medical images for telemedicine, health records, and AI diagnostics.
- Over 70% of institutions experienced data breaches, highlighting critical security gaps.
- Existing watermarking methods lack imperceptibility, robustness, and efficiency.
Purpose of the Study:
- To introduce ViTU-Net, a novel model for secure medical image watermarking.
- To address limitations in imperceptibility, robustness, and efficiency of current methods.
- To enhance data integrity, confidentiality, and authenticity of chest X-rays for pneumonia diagnosis.
Main Methods:
- Utilized a Vision Transformer (ViT) encoder and U-Net decoder with Adaptive Hierarchical Spatial Attention (AHSA).
- Employed patch-based LSB embedding in non-diagnostic regions (RONI) guided by adaptive masks.
- Integrated TuniBee Fusion optimization and cryptographic techniques (SHA-512, AES).
Main Results:
- Achieved high image quality metrics: PSNR (60.7 dB), NCC (0.9999), SSIM (1.00).
- Demonstrated ViTU-Net's resilience against various attacks through robustness analysis.
- Effectively preserved diagnostic accuracy while embedding watermarks.
Conclusions:
- ViTU-Net offers a robust and efficient solution for medical image watermarking.
- The model enhances security for AI-driven diagnostics and telemedicine applications.
- ViTU-Net ensures the integrity and confidentiality of sensitive patient data.
