Toward Medical Deepfake Detection: A Comprehensive Dataset and Novel Method

View on arXiv ← Back to list

Authors: Shuaibo Li, Zhaohu Xing, Hongqiu Wang, Pengfei Hao, Xingyu Li, Zekai Liu, Lei Zhu

Published: 2025-09-19 07:40:08+00:00

AI Summary

This paper introduces MedForensics, a large-scale dataset of AI-generated medical images across six modalities, and DSKI, a novel dual-stage deepfake detector that significantly outperforms existing methods and human experts in detecting AI-generated medical images.

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 \\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 findings

DSKI significantly outperforms existing methods and even human experts in detecting AI-generated medical images across multiple modalities. The dual-stage approach, combining CDFA and MFRM, proves crucial for achieving high accuracy. The model also demonstrates scalability by adapting to new, unseen generative models with minimal retraining.

Approach

DSKI uses a pre-trained CLIP model adapted for medical images. It employs a two-stage approach: a cross-domain fine-trace adapter (CDFA) to extract forgery clues during training and a medical forensic retrieval module (MFRM) to boost accuracy during testing via few-shot retrieval.

Datasets

MedForensics dataset encompassing six medical modalities (Ultrasound, Endoscopy, Histopathology, MRI, CT, and X-ray) and twelve state-of-the-art medical generative models.

Model(s)

A modified CLIP (Contrastive Language–Image Pre-training) model with added Cross-Domain Fine-Trace Adapter (CDFA) and Medical Forensic Retrieval Module (MFRM).

Author countries

China

← Previous