الفهرس 
Chapter 1: Introduction to Digital Image ProcessingOnly 14 pages are availabe for public view
 |  | from | 108 |  |  |
| | | from | 108 | | |
Abstract
Considering the critical nature of medical data analysis, any error in this process could potentially jeopardize patient safety or even lead to fatal consequences. In addressing this concern, our study focused on analyzing MRI brain tumors using deep learning techniques. Initially, we provided a comprehensive definition of artificial intelligence, highlighting its significance and various categories. Subsequently, we delved into the concept of machine learning, emphasizing its importance, diverse categories, key algorithms, and notable applications. Within the scope of this thesis, we specifically applied machine learning to medical diagnosis. Additionally, we briefly outlined the stages of digital image processing before delving into the intricacies of deep learning, particularly focusing on convolutional neural networks.
In this thesis, we reviewed prior studies on brain tumor classification utilizing deep learning method, focusing on various deep learning algorithms. Additionally, we introduced previous works that utilized the same database as our proposed system. Finally, we conducted comparative analyses between our proposed system and other previous studies that employed the same database.
Our objective is to develop an automated system based on Convolutional Neural Networks to classify MRI brain tumors, aiming for optimal accuracy with minimal time complexity. We varied several parameters, including the number of layers, learning rates, and image sizes using the BRATS 2015 dataset. The proposed system underwent training iteratively, employing different configurations: varying the number of layers from 6 to 10 with a step increment of 2, setting epochs to 50, and testing learning rates at 0.01, 0.1, and 0.5, along with image sizes of 64, 128, 224, and 255. Our observations indicated that as the number of layers and image size decreased, the time complexity also decreased. Additionally, we identified the best-performing system with 10 layers, achieving a higher accuracy of 99.6%.
The thesis primarily aims to theoretically explore the concept of network depth, including discussions on the features and dimensions of input images, as well as the fundamental principles of deep learning. Additionally, it practically proposes an automated system based on convolutional neural networks with optimized performance and minimized time complexity for classifying MRI brain tumors. By carefully controlling the overall network parameters and depth, the objective is to achieve the highest accuracy possible.