Scholarlink

Deepfake technology poses an emerging threat in the medical field, as its potential to manipulate medical scans could result in misdiagnoses, fraudulent claims, and serious health risks. This study addresses the urgent need for robust detection methods to safeguard medical imaging systems and patient safety. We developed a deepfake detection model by evaluating four Convolutional Neural Network (CNN) architectures: VGG16, EfficientNetV2, InceptionV3, and Sequential. These models were assessed using Precision, Recall, F1-Score, and Accuracy metrics to determine their effectiveness. CT scan images were resized to 256 \(\,\times \,\) 256 pixels, and the models were trained for 100 epochs with early stopping to enhance training efficiency. The EfficientNetV2 architecture initially achieved the highest accuracy of 92% without data augmentation. However, following data augmentation, both EfficientNetV2 and VGG16 reached an accuracy of 93%, with VGG16 being selected due to its superior recall rate of 98%, effectively minimizing false negatives and improving deepfake detection reliability.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Deepfake Detection in Cancer Medical Imaging Using CNN Architectures

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Dima Talal Alhalabi,
  • Moatsum Alawida,
  • Belkacem Chikhaoui,
  • Hala Samer Hamadeh

摘要

Deepfake technology poses an emerging threat in the medical field, as its potential to manipulate medical scans could result in misdiagnoses, fraudulent claims, and serious health risks. This study addresses the urgent need for robust detection methods to safeguard medical imaging systems and patient safety. We developed a deepfake detection model by evaluating four Convolutional Neural Network (CNN) architectures: VGG16, EfficientNetV2, InceptionV3, and Sequential. These models were assessed using Precision, Recall, F1-Score, and Accuracy metrics to determine their effectiveness. CT scan images were resized to 256 \(\,\times \,\) 256 pixels, and the models were trained for 100 epochs with early stopping to enhance training efficiency. The EfficientNetV2 architecture initially achieved the highest accuracy of 92% without data augmentation. However, following data augmentation, both EfficientNetV2 and VGG16 reached an accuracy of 93%, with VGG16 being selected due to its superior recall rate of 98%, effectively minimizing false negatives and improving deepfake detection reliability.