Deep Transfer Learning Networks for Brain Tumor Detection: The Effect of MRI Patient Image Augmentation Methods

Abstract

The exponential growth of deep learning networks has enabled us to handle difficult tasks, even in the complex field of medicine with small datasets. In the sphere of treatment, they are particularly significant. To identify brain tumors, this research examines how three deep learning networks are affected by conventional data augmentation methods, including MobileNetV2, VGG19, and DenseNet201. The findings showed that before and after utilizing approaches, picture augmentation schemes significantly affected the networks. The accuracy of MobileNetV2, which was originally 85.33%, was then enhanced to 96.88%. The accuracy of VGG19, which was 77.33%, was then enhanced to 95.31%, and DenseNet201, which was originally 82.66%, was then enhanced to 93.75%. The models' accuracy percentage engagement change is 13.53%, 23.25%, and 23.25%, respectively. Finally, the conclusion showed that applying data augmentation approaches improves performance, producing models far better than those trained from scratch.

The impact of image augmentation techniques of MRI patients in deep transfer learning networks for brain tumor detection

“…With the help of their investigation, the researchers were successful in obtaining a 92.34 percent F1 detection score with the ResNet50 network. Abdalla et al [18] investigated typical data augmentation methods' effects on three deep learning networks, namely MobileNetV2, VGG19, and DenseNet201, to detect brain tumors. The results demonstrated that picture augmentation schemes had a significant impact on the networks both before and after using approaches.…”