ML Security PapersToward Medical Deepfake Detection: A Comprehensive Dataset and Novel Method
Shuaibo Li 1, Zhaohu Xing 1, Hongqiu Wang 1, Pengfei Hao 1, Xingyu Li 1, Zekai Liu 1, Lei Zhu 1,2
Published on arXiv
2509.15711
Output Integrity Attack
OWASP ML Top 10 — ML09
Key Finding
DSKI significantly outperforms both existing media forensics methods and human medical experts at detecting AI-generated medical images across multiple modalities
DSKI (Dual-Stage Knowledge Infusing detector)
Novel technique introduced
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 \textbf{MedForensics}, a large-scale medical forensics dataset encompassing six medical modalities and twelve state-of-the-art medical generative models. We also propose \textbf{DSKI}, a novel \textbf{D}ual-\textbf{S}tage \textbf{K}nowledge \textbf{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.
Key Contributions
- MedForensics: large-scale medical forensics dataset with ~58,000 image pairs across 6 modalities (ultrasound, MRI, CT, X-ray, endoscope, histopathology) and 12 state-of-the-art generative models
- DSKI detector with cross-domain fine-trace adapter (CDFA) extracting subtle forgery clues from both spatial and noise domains using multi-scale inception modules and constrained CNNs
- Medical Forensic Retrieval Module (MFRM) enabling few-shot retrieval-augmented detection at test time to boost accuracy and scalability
🛡️ Threat Analysis
Directly addresses AI-generated content detection — identifying synthetic/deepfake medical images produced by diffusion models and GANs. The paper proposes a novel forensic detection architecture (DSKI) with purpose-built components (CDFA, MFRM) and a new dataset (MedForensics), qualifying as a novel forensic technique rather than a mere domain application of existing detectors.