SYSTEMATIC REVIEW: DEEP FAKE DETECTION IN MEDICAL IMAGES

Abstract

 The rapid advancement of generative models, particularly Generative Adversarial Networks (GANs), has led to the rise of highly realistic synthetic content, including medical deep fakes. Although such technologies offer significant potential for applications in medical training, simulation, and image improvement, they raise critical concerns regarding the authenticity, security, and reliability of medical data. Despite growing awareness of these threats, there is currently no comprehensive systematic review focusing exclusively on deep fake detection techniques within the domain of medical imaging. To address this gap, the present study presents the first known systematic literature review (SLR) on medical deep fake detection in images. The review was conducted through a structured and methodical examination of top-tier scientific databases, including IEEE Xplore, Elsevier ScienceDirect, SpringerLink, ACM Digital Library, and arXiv. Following a rigorous selection process based on relevance, novelty, and contribution, thirty peer-reviewed studies were analyzed. The main contributions of this work are as follows: (1) a comparative analysis of traditional machine learning and deep learning-based detection techniques such as ResNet, DenseNet, CNNs, YOLO architectures, and ensemble frameworks; (2) benchmarking of detection models against real-world datasets including CT-GAN, DFDC, and LIDC-IDRI; (3) assessment of core evaluation metrics such as accuracy, precision, recall, AUC, and model robustness; (4) identification of ethical concerns surrounding privacy, potential misuse, and the absence of clinical validation standards; and (5) recommendations for future improvements in model robustness, real-time operability, and interpretability. This review aims to serve as a foundational reference for researchers in the field of medical deep fake detection by highlighting current advancements, unresolved challenges, and prospective research directions.