الفهرس 
Introduction and Literature ReviewOnly 14 pages are availabe for public view
 |  | from | 107 |  |  |
| | | from | 107 | | |
Abstract
Among the organs in the human body, the human brain is considered one of the most important and complex organs. It contains most of the soft tissues and nerve cells that help the human to carry out the functions and activities normally. The brain tumor is considered the most life-threatening and aggressive disease that affect the tissues of human brain, especially Central Nervous System (CNS), whether it is for adults or children. Usually, brain tumors are uncontrolled and may lead to death if the patient does not receive appropriate treatment according to the type of tumor. In 2021, approximately 83,570 patients are diagnosed with brain tumor in the USA and approximately 18,600 patients died from this disease. The brain tumors can be divided into two main categories; benign (noncancerous) and malignant (cancerous) tumors. The benign tumors are formed when cells duplicate more than they ought to or do not die when they ought to. Therefore, this type of tumor may press on some parts of the brain. The benign tumors spread very slowly and cannot grow to other parts of the human body so it is considered one of the least dangerous types of tumors and do not require a critical treatment. On the other hand, the spread of malignant tumor is very quick, and attack other normal (healthy) cells leading to its grow in other parts of the human body. Due to the criticality of the brain tumors and their significant impact on human life, the patient must receive appropriate treatment. To perform this task accurately, it is necessary to determine the type of tumor, as each type has a specific treatment protocol. Then, the attention are turned to use the artificial intelligence (AI) and its subtypes of machine learning (ML) and deep learning (DL) to detect the type of tumor which helps to reduce the risk of misdiagnosis for physicians. DL is a subfield of ML and considered as one of remarkable computational intelligence techniques. DL algorithms have a huge employment in various fields and have a significant success for medical imaging classification and automatic diagnosis. Fine-tuning process takes place through replacing and retaining the pre-trained network on the target dataset to apply back-propagation. Then the target dataset may be classified through the last fully connected layer. However, in the other way, the last fully connected layers are removed, because they act as features passed to a classifier such as SVM, KNN, Naïve Bayes, etc. to complete classification process. Chapter 1 is the introduction about the brain anatomy, functions and medical problems. Then, the role of the artificial indigence in medical imaging has been discussed. The relevant literature review has been explored to indicate the importance of our study and to show the progress of the deep learning techniques for diagnosing the brain tumors. Chapter 2 explored the physical principles of medical diagnostic imaging, whereas four modalities are presented with their physical way of working in a nutshell, namely X-rays for diagnosis and treatment, CT, Ultrasound, and then the MRI device was explained and some important details were entered into it because it is essential in our work. Chapter 3 offers the basics of the image processing using the machine learning and deep learning techniques. It starts with the different stages of preprocessing of medical images. Then followed by the definitions of the segmentation tools and the features extraction to understand the identification process. Chapter 4 presents hybrid convolutional neural networks for classifying the brain tumor classes based on MRI scans. The main goal of the work is to enhance the performance of CNNs for detecting and classifying the brain tumors categories based on MRI scans. A database compiled from three other databases (Figshare, SARTAJ, and Br35h) are utilized to perform this task. The total number of images in our database is 2880 scans, including the four main types of brain tumors (no tumor, glioma tumor, meningioma tumor and pituitary tumor). Four refined CNNs are proposed for classifying the brain tumors categories in the target dataset. These networks are tested using new dataset. The confusion matrix is utilized for estimating the testing and validation accuracy of these networks, and achieved higher validation accuracy.