الفهرس 
Generative Adversarial Network (GAN) in Data AugmentationOnly 14 pages are availabe for public view
 |  | from | 172 |  |  |
| | | from | 172 | | |
Abstract
The wide prevalence of brain tumors in all age groups necessitates having the ability to make an early and accurate identification of the tumor type and thus select the most appropriate treatment plans that yield the lowest mortality.The application of deep neural networks(DNN to radiology has aided radiologists to classify and segment the type of brain tumor tissues from magnetic resonance images(MRI)more accurately. In this thesis, we introduce an integrated system that consists of two phases to classify and segment different types of brain tumors tissues.
Patient privacy and MRIs collection costs are causing a suboptimal number of dataset MRIs number. So, deep learning models may suffer from underfitting which therefore reflects as unacceptable system accuracy. In the classification phase, to solve this problem, the progressive growing generative adversarial network (PGGAN) augmentation method is used for optimizing the learning stage which maximizing the overall efficiency of the system by increasing the dataset size by balancing each class number of samples. The objective of current research on tumor classification is to increase the precision of existing classification techniques by increasing the dataset MRI number. In the classification phase, three deep learning models: Visual Geometry group (VGG)19+Convolutional Neural Network (CNN), VGG19+Gated Recurrent Units (GRU), and VGG19+Bidirectional Gated Recurrent Units (Bi-GRU)) model based on VGG19 model as a feature extractor, are introduced. In addition to using classical augmentation techniques with these three models, PGGAN augmentation model is used to produce ‘realistic’ MRIs of brain tumors and to help overcoming the shortage of images needed for deep learning. Results indicated the ability of our models to classify brain tumors more accurately than previous studies. Other performance metrics were also measured including precision, recall, specificity, negative predictive value, and Matthew’s correlation coefficient range.