Hybrid Deep Learning for Detecting Fake Medical Images and Heart Sounds.

The rapid development of deepfake technology has raised serious concerns in numerous fields, particularly in healthcare, where manipulating medical data could compromise patient safety and undermine the reliability of diagnostic processes. This study presents innovative methodology for identifying deepfake alterations in medical images and audio, focusing on MRI and CT scans and heart sound recordings. The proposed system integrates advanced image and audio processing techniques to detect synthetic modifications. Medical images undergo pre-processing steps such as resizing, grayscale conversion, noise injection, and data augmentation, while heart sounds are examined using feature extraction techniques like the Zero Crossing Rate. To detect manipulated regions in images, YOLOv8 is employed, followed by classification through deep-learning models like VGG-19 and ResNet-50. Important metrics are implemented to assess the system’s performance, namely accuracy 99.7, precision 86.6, recall 92.8, F1-score 89.6, error rate 0.215, and confusion matrix, ensuring comprehensive assessment of its reliability. Outcomes highlight effectiveness of this approach in identifying deepfake medical data with high accuracy 99.7, offering a vital solution for maintaining the integrity of healthcare diagnostics in the face of evolving threats from synthetic data manipulation.