Back-in-Time Diffusion: Unsupervised Detection of Medical Deepfakes

View on arXiv ← Back to list

Authors: Fred Grabovski, Lior Yasur, Guy Amit, Yisroel Mirsky

Published: 2024-07-21 13:58:43+00:00

AI Summary

This paper introduces Back-in-Time Diffusion (BTD), a novel unsupervised medical deepfake detector using diffusion models. Unlike existing methods, BTD reverses the diffusion process on a suspected image without adding noise, resulting in a residual containing only anomalous forensic content, significantly improving detection accuracy.

Abstract

Recent progress in generative models has made it easier for a wide audience to edit and create image content, raising concerns about the proliferation of deepfakes, especially in healthcare. Despite the availability of numerous techniques for detecting manipulated images captured by conventional cameras, their applicability to medical images is limited. This limitation stems from the distinctive forensic characteristics of medical images, a result of their imaging process. In this work we propose a novel anomaly detector for medical imagery based on diffusion models. Normally, diffusion models are used to generate images. However, we show how a similar process can be used to detect synthetic content by making a model reverse the diffusion on a suspected image. We evaluate our method on the task of detecting fake tumors injected and removed from CT and MRI scans. Our method significantly outperforms other state of the art unsupervised detectors with an increased AUC of 0.9 from 0.79 for injection and of 0.96 from 0.91 for removal on average. We also explore our hypothesis using AI explainability tools and publish our code and new medical deepfake datasets to encourage further research into this domain.

Key findings

BTD significantly outperforms state-of-the-art unsupervised detectors, achieving an AUC of 0.9 for injection and 0.96 for removal deepfakes on average. The method shows robustness across different medical imaging devices and deepfake generation techniques, except for CT-GAN removal attacks where all methods performed poorly.

Approach

BTD trains a diffusion model on authentic medical images. For detection, it performs a single backward diffusion step on a suspected image without adding noise. The residual between the original and denoised image is used to score the likelihood of a deepfake.

Datasets

Duke Breast Cancer MRI dataset, LIDC dataset, and six novel medical deepfake datasets (CTGAN-CT-Inject, CTGAN-CT-Remove, SD-CT-Inject, SD-CT-Remove, SD-MRI-Inject, and SD-MRI-Remove) created using CT-GAN and Stable Diffusion.

Model(s)

U-Net based Denoising Diffusion Probabilistic Model (DDPM)

Author countries

Israel

← Previous