Secure breast cancer imaging: A novel advanced generative model encryption approach

The proposed work presents a novel Artificial Intelligence (AI) technique, specifically a Cyclic-Generative Adversarial Network (Cyclic GAN)-based encryption framework, for securing breast cancer medical images, leveraging AI's ability to learn from unpaired datasets. Unlike traditional chaotic...

Celý popis

Uloženo v:
Podrobná bibliografie
Vydáno v:Engineering applications of artificial intelligence Ročník 163; s. 112747
Hlavní autoři: Inam, Saba, Kanwal, Shamsa, Hajjej, Fahima, Saleh Alluhaidan, Ala
Médium: Journal Article
Jazyk:angličtina
Vydáno: Elsevier Ltd 01.01.2026
Témata:
Artificial intelligence
Cycle generative adversarial network
Deep learning
Discriminator
Generator
Image reconstruction
Deep learning
Cycle generative adversarial network
Artificial intelligence
Image reconstruction
Discriminator
Generator
ISSN:0952-1976
On-line přístup:Získat plný text
Tagy: Přidat tag
Žádné tagy, Buďte první, kdo vytvoří štítek k tomuto záznamu!
Popis
Shrnutí:The proposed work presents a novel Artificial Intelligence (AI) technique, specifically a Cyclic-Generative Adversarial Network (Cyclic GAN)-based encryption framework, for securing breast cancer medical images, leveraging AI's ability to learn from unpaired datasets. Unlike traditional chaotic encryption methods requiring aligned input-output pairs, this proposal employs dual generators and discriminators to transform images while preserving features, ensuring secure and high-quality encryption. The model is optimized using binary cross-entropy loss for precise adversarial training and evaluated on breast cancer datasets. The integration of a Structural Similarity Index Metric (SSIM) optimized loss function of 0.9944 (Surpassing conventional chaotic encryption by 21.3 %) ensures exceptional preservation of structural integrity, brightness, and contrast in decrypted images. Quantitative evaluation demonstrates outstanding reconstruction quality with Peak Signal-to-Noise Ratio (PSNR) values reaching 39.08 dB (exceeding wavelet-based methods by 41 %). A comprehensive security analysis confirms the system's resistance to statistical and discriminative attacks while maintaining efficient encryption and decryption performance. The system's dual generator-discriminator architecture optimized with Binary Cross Entropy loss provides following three key advantages over state-of-art techniques: 40 % faster encryption/decryption than pixel-scrambling methods, 92 % resistance to statistical attacks compared to 78 % for Advanced Encryption Standard based approaches elimination of the paired data requirement that hinders conventional deep learning solutions. The Number of Pixel Change Rate (NPCR) > 99.2 % and Unified Average Changing Intensity (UACI) ≈ 33.4 % confirm protection against differential attacks.
ISSN:0952-1976
DOI:10.1016/j.engappai.2025.112747