Brain Tumor Detection with Deep Learning Methods’ Classifier Optimization Using Medical Images

Abstract

It is known that, with the development of artificial intelligence science in recent years, it has started to be used in all areas of life. Due to the increase in diseases that threaten human life, such as epidemics and cancer, more attention has been paid to research in this field. Especially in the field of biomedical image processing, very successful results have been obtained in recent years with the use of deep learning methods. For this study, MR images are utilized to diagnose brain tumors. To assist doctors and radiologists in automatic brain tumor diagnosis and to overcome the need for manual diagnosis, a brain MR image automated classification system is being developed. The data used in the study are open access data obtained from the Kaggle library. This paper presents a novel approach for classifying brain MR images utilizing a dataset of 7022 MR images. To give an unbiased evaluation of the dataset, it is divided into a 40% test and 60% training set. Respectively, VGG, ResNet, DenseNet and SqueezeNet architectures are trained and used for feature extraction from brain MRI images. In order to classify the extracted features, machine learning methods (Support Vector Machines, K-Nearest Neighbors, Naive Bayes, Decision Tree, Linear Regression Analysis) are applied first, then an ensemble learning method is applied and the best validation method is selected. In addition, parameter optimization is applied to the trained CNN algorithms. In order to develop the proposed methods, the Python software program was used in the training and testing phases of the models, and the classification success rates were mutually evaluated. Among the results found, it can see that the ResNet architecture reached 100% accuracy. The data obtained as a result of the study were compared with the results of similar studies. In conclusion, the techniques and methods applied highlight their effectiveness in accurately classifying brain MRI images and their potential to improve diagnostic capabilities.