Thesis (MSc) - Cyprus International Relations. Institute of Graduate Studies and Research Computer Engineering Department

Includes bibliography (sheets 44-45)

ABSTRACT
The brain or spinal cord is where a primary brain tumor or spinal cord tumor originates. Seventy eight thousand (78,000) people will be discovered with main brain and CNS illnesses this year. This year, 23,770 individuals in the United States (13,000 males and 10,000 women) will be discovered with a large malignant brain tumor. About 4,000 young individuals and children will be discovered with a CNS or brain tumor this year. There are secondary brain tumors, often known as brain metastases, in addition to primary brain tumors. When a tumor appears in another territory of the body and move to the brain, this is called metastasis. Because brain tumors are the most nonstop or frequent and violent cancer, they have a relatively short life suspense. As a result, therapy planning is a crucial step in enhancing the quality of life of patients. Typically, CT, MRI, and ultrasound scans are utilized to identify malignancies in brain, lungs, liver, breast, prostate, and other organs. MRI images are utilized to diagnose brain malignancies in particular. Large volumes of data collected by MRI scans, on the other hand, make it impossible to manually distinguish tumors from non-tumors at a given moment. However, there are significant drawbacks (for example, reliable quantitative measurement is only available for a small number of images). To reduce human mortality, reliable and automated categorization techniques are required. The vast and structural changes of the region around the brain tumor make automated categorization of brain tumors a tough endeavor. A robot or computer must be able to recognize what it sees in order to function. The computer must be able to classify what a picture depicts just by looking at it. This is a simple task for people, but it is not that simple for computers. To categorize a picture, the computer must go through numerous processes.
In this proposed system we present a comparative study between conventional machine learning models i.e. SVM and KNN classifiers and deep learning model. In start he input MRI images are preprocessed to remove noise & make the image suitable for the next steps. After completion of pre-processing, properties are bring out from the images using HOG and LBP and then given to SVM and KNN. At the end images are classified through trained KNN and SVM into normal and tumor class. Then we trained a CNN model on the same dataset which is used for conventional machine learning classifier and then compare the results of conventional machine learning and CNN model. The conventional machine learning classifiers achieved accuracy for KNN is 85% and SVM is 96% while deep learning model achieved the highest accuracy of 98%.
Keywords: Brain Tumor, CNN model, HOG, LBP, Machine Learning, SVM and KNN Classifiers.