Deep learning steganography for big data security using squeeze and excitation with inception architectures

Area of Science:

• Computer Science
• Medical Imaging
• Cybersecurity

Background:

• Secure transmission of sensitive medical data is crucial due to big data growth in telemedicine and digital forensics.
• Conventional steganography methods struggle with diagnostic integrity and robustness against noise and transformations.

Purpose of the Study:

• To develop a novel deep learning-based steganographic framework for secure medical image transmission.
• To address limitations of conventional methods in maintaining diagnostic integrity and robustness.

Main Methods:

• Proposed a framework combining Squeeze-Excitation (SE) blocks, Inception modules, and residual connections.
• Encoder uses dilated convolutions and SE attention for embedding secret medical images into cover images.
• Decoder utilizes residual and multi-scale Inception-based feature extraction for reconstruction.

Main Results:

• The model achieves high Peak Signal-to-Noise Ratio (PSNR) values (39.02, 38.75) and Structural Similarity Index (SSIM) values (0.9757) on MRI and OCT datasets.
• Demonstrates minimal visual distortion, confirming the efficacy of the steganographic approach.
• Designed for real-time, low-power deployment on NVIDIA Jetson TX2 for edge healthcare applications.

Conclusions:

• The developed deep learning framework offers a secure, high-capacity steganographic solution for privacy-sensitive environments.
• The model's real-time performance and robustness make it suitable for practical applications in telemedicine and digital forensics.
• This research advances secure medical data handling, preserving both confidentiality and diagnostic quality.