الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 98 |  |  |
| | | from | 98 | | |
Abstract
Nowadays, medical imaging is one of the most sophisticated and emerging fields. In every hospital, use some devices to diagnose the patient’s condition, such as MRI, CT scan, ultra scan, X-rays, etc. All of these devices produce images. Based on the information provided by the images, the doctor makes a diagnosis of the patient’s condition. Pre-processing of these images is very important for radiologists and doctors to diagnose the patient correctly.
Brain tumor is one of the most serious problems in medicine. To diagnose the tumor in its early stage, a correct diagnosis is needed. There are so many devices that produce images to diagnose brain tumors, including CT (computed tomography), MRI (magnetic resonance imaging), etc. In this dissertation, the detection of brain tumors in MRI images is presented to enable the proper visualization and diagnosis of patients and to facilitate radiologists and physicians. In this work, the analysis of normal MR images and tumor detection images is also performed.
A comparative study of three methods is performed to evaluate their relative performance in segregating brain tumors. These methods are Seeded Region Growing, K-means, and Global Thresholding. Median and Soft Weighted Median filters are used before the segmentation algorithms to remove any noise from the images. The filters have improved the accuracy of image segmentation. Various images are obtained from the Cancer Imaging Archive (TCIA) and Kaggle are used. The experimental results show that the K-means method provides better accuracy than Seeded Region Growing method and Global Thresholding method.
We demonstrate a way to segment the brain tumor using modify U-net net-works. The U-net is further developed by adding a convolutional layer and changing the filter size. The modified U-net is compared with traditional U-net, K-means, and Thresholding methods. The experiments are carried out using three multimodal brain tumor image segmentation datasets (FIGSHARE database), which contain 3929 abnormal (with a tumor) and normal brain MRI images, 100 images obtained from “The Cancer Imaging Archive (TCIA)” and BraTS 2019 challenge which include 4600 cases (normal and abnormal) of HGG (high-grade glioma) and LGG (low-grade glioma). The modified U-net achieves higher accuracy than other methods.
The Shuffle and Input Normalization are used as preprocessing methods. Also, the Batch Normalization layer is used with the modified U-net. The hidden layers are increased in the U-net. The experimental results show an improvement in the segmentation accuracy.