Deepfake Detection in Medical Images Using Mask RCNN

Deepfakes in medical imaging present a significant threat to the integrity of healthcare systems by enabling the malicious manipulation of diagnostic scans. This study introduces a novel framework for detecting deepfake alterations in brain MRI images using a Mask R-CNN–based approach. The proposed system classifies images into four distinct categories: (i) real tumor, (ii) no tumor, (iii) deepfake-added tumor, and (iv) deepfake-removed tumor. Leveraging pixel-level instance segmentation, the model effectively captures subtle modifications that are often indiscernible to the human eye. The overall system architecture integrates a Next.js–based frontend with a Python backend, where the Mask R-CNN model is trained on a curated dataset comprising 1,300 brain MRI scans. Experimental evaluations demonstrate that the proposed method accurately localizes and classifies manipulated regions, thereby enhancing diagnostic trustworthiness and safeguarding medical data authenticity. This research contributes to the growing field of medical deepfake detection by combining segmentation and classification techniques to achieve robust and reliable detection performance. Future work will focus on improving model generalizability and scalability across diverse imaging modalities and clinical settings.