.png){kind=icon}
.png){kind=icon}
Return
Integrating Fourier analysis and deep learning for robust detection of deep fake brain magnetic resonance images
.png){kind=icon}
PRE
AI
V
Y
P
P
D
P
R
A
P
DOI：10.1016/j.patrec.2026.03.016.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Recent advances in generative models have enabled the creation of highly realistic synthetic medical images, helping to address the challenges of data scarcity. However, these technologies pose a threat by generating realistic fake medical images that can mislead clinical diagnoses. In this work, we proposed a framework for detecting fake brain MRIs using handcrafted features and deep representations. Specifically, we extracted raw intensities, Error Level Analysis (ELA), wavelet, gradient, and Fourier features, along with embeddings from a pre-trained VGG19 network. We evaluated these features, derived from both real and fake MRIs, using machine learning (K-Means clustering, Support Vector Machines (SVM)) and deep learning (MobileNetV2) approaches. Experimental results show that Fourier-based features achieve the highest detection accuracy of 99.5% with SVM and 99.8% with CNNs. Additionally, VGG19 embeddings achieved 98.8% accuracy with SVM even in low-data regimes. These findings highlight the effectiveness of combining domain-aligned features with supervised learning for robust detection of medical deepfakes.
Keywords:
Medical deep fake detection
Synthetic data
Fourier features
Deep feature embeddings
Machine learning
Deep learning
Journal
IF:
3.3
Papers:
7.8K
Citations:
1.6W
Organization
Cited Papers