Brain tumor diagnosis and classification using deep learning methods

Deep learning is a field of study that includes Artificial Neural Networks (ANN) and similar machine learning algorithms that contain one or more hidden layers. In other words, the computer obtains new data from the available data by using at least one artificial neural network and many algorithms. With these studies, deep learning achieves success in managing and directing life correctly in many areas. Deep learning used in the medical field provides great support to both patients and doctors by detecting brain tumors quickly and accurately. However, the inadequacy of existing solutions and technological methods, the scarcity of dataset content in some areas, the lack of sufficient diversity, and the high noise levels in the relevant images make it difficult to develop successful applications in these areas. Moreover, the number of patients per doctor, especially in underdeveloped countries, causes doctors to suffer from stress, loss of attention, fatigue, etc. from time to time. For these reasons, incorrect or incomplete evaluations in the diagnosis of brain tumors negatively affect human health. This study, conducted to eliminate the negativities and support doctors, focuses on the classification of brain tumors using convolutional neural networks (CNN) models. Brain tumor images are classified on Google Colab using deep learning algorithms RESNET50, ALEXNET, YOLOV8 and the CNN model proposed in this thesis. On Google Colab, 4 different models, RESNET50, ALEXNET and the CNN model proposed in this study, are determined by the classification method, and in the variants of the YOLOV8 model, the model with the highest learning rate is determined and compared by the object detection method. Magnetic Resonance (MR) images used as the dataset in this study contain a total of 7023 images, consisting of four classes. The images in the content of the data set are divided into two groups: training and testing, and the content of each group consists of four classes. These groups are glioma tumor, meningioma tumor, no tumor, and pituitary tumor, respectively. The data set was taken from the public Kaggle database, but due to the different sizes of the MR image size it was rearranged and the size of each image was standardized to 300x300. In this study, which is carried out to eliminate the negativities and support doctors, four deep learning algorithms are used on Google Colab, and the model with the highest accuracy rate is tried to be obtained among all the results obtained. Different models have been used on the same data set in the literature, but no comparison has been made using four different models on Google Colab. This study attempted to eliminate this deficiency. In this research, it was aimed at diagnosing and classify brain tumors with high accuracy by using different models with deep learning methods. RESNET50, ALEXNET, our CNN model, and YOLOV8 are used in brain tumor diagnosis and classification. Different success rates were obtained in each model. The suitability of the data set used in the model was effective in obtaining different success rates. Keywords: Brain tumor, Deep learning, Classification, Cnn, Alexnet, Resnet50, Yolov8.