Peshraw Ahmed Abdalla

The exponential growth of deep learning networks has enabled us to handle difficult tasks, even in the complex field of medicine. Nevertheless, for these models to be extremely generalizable and perform well, they need to be applied to a vast corpus of data. In order to train transfer learning networks with limited datasets, data augmentation techniques are frequently used due to the difficulties in getting data. The use of these methods is crucial in the medical industry in order to enhance the number of cancer-related magnetic resonance imaging pathology scans. This study evaluates the results of data augmentation methods on three deep transfer learning networks, such as InceptionV3, VGG16, and DenseNet169, for brain tumor identification. To demonstrate how data augmentation approaches affect the performance of the models, networks were trained both before and after the application of these methods. The outcomes revealed that the image augmentation strategies have a big impact on the networks before and after using techniques, such as the accuracy of VGG16 is 77.33% enhanced to 96.88%, and InceptionV3 changed from 86.66 to 98.44%, and DenseNet169 changed from 85.33 to 96.88% the accuracy percentage increase of the models are 19.55%, 11.78%, and 11.55%, respectively.