Toward Medical Deepfake Detection: A Comprehensive Dataset and Novel Method

2025-09-19

AI Summary

AI-generated medical images pose serious security risks—including diagnostic deception and insurance fraud—yet existing media forensics methods, designed for natural images, fail to detect subtle, modality-specific artifacts in medical imagery and lack dedicated benchmark datasets. Method: We introduce MedForensics, the first large-scale medical image forensics dataset, covering six imaging modalities and twelve state-of-the-art generative models. We further propose DSKI (Dual-Stage Knowledge Injection detector), a novel forensic framework integrating cross-domain fine-grained adapters—modeling spatial and noise-domain forgery traces—and a medical forensics retrieval module that jointly leverages vision-language features and few-shot retrieval. Contribution/Results: Extensive experiments demonstrate that DSKI significantly outperforms existing SOTA methods and even human experts across multimodal medical images. It exhibits strong generalization to unseen models and modalities, and shows promising potential for clinical deployment.

Technology Category

• Computer Vision: Medical and Biological Imaging
• Data Mining & Knowledge Management: Mining of Visual, Multimedia & Multimodal Data
• Machine Learning: Multimodal Learning

Application Category

• Security and Privacy: Data transparency and provenance
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Web Mining and Content Analysis: Large pretrained models with web data

Abstract

The rapid advancement of generative AI in medical imaging has introduced both significant opportunities and serious challenges, especially the risk that fake medical images could undermine healthcare systems. These synthetic images pose serious risks, such as diagnostic deception, financial fraud, and misinformation. However, research on medical forensics to counter these threats remains limited, and there is a critical lack of comprehensive datasets specifically tailored for this field. Additionally, existing media forensic methods, which are primarily designed for natural or facial images, are inadequate for capturing the distinct characteristics and subtle artifacts of AI-generated medical images. To tackle these challenges, we introduce extbf{MedForensics}, a large-scale medical forensics dataset encompassing six medical modalities and twelve state-of-the-art medical generative models. We also propose extbf{DSKI}, a novel extbf{D}ual- extbf{S}tage extbf{K}nowledge extbf{I}nfusing detector that constructs a vision-language feature space tailored for the detection of AI-generated medical images. DSKI comprises two core components: 1) a cross-domain fine-trace adapter (CDFA) for extracting subtle forgery clues from both spatial and noise domains during training, and 2) a medical forensic retrieval module (MFRM) that boosts detection accuracy through few-shot retrieval during testing. Experimental results demonstrate that DSKI significantly outperforms both existing methods and human experts, achieving superior accuracy across multiple medical modalities.

Problem

• Detecting AI-generated fake medical images
• Lack of specialized datasets for medical forensics
• Inadequate existing methods for medical image characteristics

Innovation

• Large-scale medical forensics dataset
• Dual-stage knowledge infusing detector
• Cross-domain fine-trace adapter extraction