Detection and classification of brain tumors from MR images based on deep learning

The number of deaths stemming from brain tumors is increasing every day. Brain tumors grow quite fast. Early diagnosis has a vital role in diagnosis of cancer, treatment planning and evaluation of treatment. If a patient with brain tumor has not received accurate and early treatment, the chance of survival decreases and this may cause death. Medical imaging plays a crucial role in detection and diagnosis of brain tumors. Magnetic Resonance Imaging (MRI) is one of the most popular medical imaging methods. Detection of tumors and description of tumor features are done by experts through MRI. Defining the quality of brain tumors depends on doctors' experience and knowledge. World Health Organization (WHO) reports show that the number of people suffering from brain tumors is increasing significantly every year. Tumors have different shape and size and may exist in different parts of the brain. This creates a challenge for experts to detect the tumor. Detection of tumors by experts is a long and sensitive process. Experts' experience affects the success of tumor detection process. Manual detection of tumors is not an effective method in cases where the number of patients is redundant. Therefore, there is a need for detection of tumors automatically. Improving diagnosis skills of doctors and diminishing the time spent by experts for accurate diagnosis is possible through Computer Assisted Automatic Detection Systems (CAADS). In this dissertation study, 5 different CAADS, which can automatically identify tumors and classify them according to tumor levels and tumor types through brain MR images, have been designed. The performance of CAADS, which were designed using 6 different databases, has been evaluated. Two of the CAADS covers the pretreatment, morphological operations, side detection and definition of the textural, statistical, morphological and color qualities of brain tumors and their classification. Other three systems, on the other hand, are based on deep learning architectures such as Convolutionary Nerve Networks (CNN), Local Convolutionary Nerve Networks (LCNN), AlexNet, VGG16, Local Reciever Areas and Excessive Learning Machine. Advantages and challenges of each CAADS have been thoroughly examined in this study. As a result of studies, it is seen that systems successfully identify and classify brain tumors.