A Lightweight Deep Learning Framework for Real-time Detection of Deepfake Medical Images

Medical images require authentic verification for accurate diagnosis and treatment planning. The increasing availability of image manipulation tools poses significant challenges and threats to medical imaging. This paper presents a lightweight, deep-learning-based framework for the detection of deepfake medical images. The proposed model integrates the MobileNetV2 architecture with an efficient channel attention (ECA) module to enhance performance. The ECA module was integrated after the expansion layers of the MobileNetV2 network to improve the feature extraction by adjusting the channel-wise representations. This enhanced the ability of the proposed model to capture fine-grained details and subtle inconsistencies. The suggested method was assessed using the CT-GAN benchmark dataset. Results from the experiments showed that this proposed method achieved an accuracy of 99.81 and a precision of 99.80. These results demonstrate that the proposed model can easily detect small modifications in the medical images. The lightweight design makes the proposed approach suitable for real-world medical image analysis and ensures the authenticity of medical images.