Cybersecurity for sensitive image encryption in untrusted environment based on neural cryptography and deep autoencoders: A systematic review
Faiez Musa Lahmood Al-Rufaye, Alaa Kadhim Farhan, Nagham Hamid Abdul Mahdi · Array · 2026
In untrusted system contexts like cloud, edge, and IoT, there is a growing reliance on sensitive images, particularly medical images, with significant issues for data integrity, confidentiality, and key management throughout transmission and storage. With a focus on neural and automatic cipher-based strategies for securing images in untrusted networks, this systematic review attempts to summarize recent work published in the literature (2022–2025) in order to understand the available research and trends for protecting sensitive images. The selection process for this study was done in accordance with PRISMA 2020 guidelines. The search yielded 221 records. After removing 59 duplicates, 162 records remained for title and abstract screening. 27 records were excluded for irrelevance, and for 135 reports, the full text was sought. Forty-three reports could not be retrieved due to access restrictions, and all 92 full-text reports were evaluated without further exclusions (0), yielding a final synthesis of 92 full-text studies. The results suggest that the findings constitute “full-text evidence,” but that unretrieved reports produce availability bias that should be explicitly mentioned by PRISMA. Additionally, the synthetic reading suggests that the analysis should not focus solely on the statistics of image indicators to ensure that a comprehensive threat model and relevant key management in untrusted environments are included in any analysis of the effectiveness of proposed solutions.