Transfer Learning-Based MobileNetV2 Framework for Medical Deepfake Detection Using Chest X-Ray Images

Abstract

Medical deepfakes are an emerging cybersecurity threat to healthcare systems because image-manipulated radiology can influence the interpretation of the diagnostic results and clinical decision making. The artificial intelligence-generated or modified medical images may conceal diseases, introduce novel abnormalities, or endanger patient safety with false graphical data. Consequently, more believable medical deepfake detection systems are required to provide trustful healthcare imaging environments. The paper shows a transfer learning based lightweight convolutional neural network-based framework to detect manipulated chest X ray images. The experimental data was built on the basis of publicly available pneumonia chest X ray images in FAKE and REAL categories. Artificial manipulations of a binary classification task to produce false medical images include blur, noise addition, brightness addition, and compression changes. The proposed framework used the MobileNetV2 framework and the pretrained ImageNet weights to learn effective feature extraction and classification. The testing accuracy was determined to be approximately 98 percent by using 720 chest X ray images in experimental testing. The findings show that lightweight transfer learning models are applicable in successfully verifying medical images and improving clinical confidence in medical imaging systems.