Detection of brain tumor type using convolutional neural networks

The brain, which controls important vital functions such as vision, hearing and movement, negatively affects our lives when it is sick. Of these diseases, the deadliest is undoubtedly the brain tumor, which can occur in all age groups and can be benign or malignant. Therefore, early diagnosis and prognosis are very important. MR (Magnetic Resonance) images are used for the detection and treatment of brain tumor types. Successful results in the detection of diseases from medical images with Convolutional Neural Networks (ESA) depend on the optimum creation of the number of layers and other hyper-parameters. In this study, we propose an ESA model that will achieve the highest accuracy with the least number of layers. A dataset consisting of four different classes (Meningioma, Glioma, Pituitary, and Normal) was used for the training of ESA models. After training and testing on 50 different deep learning models with variations in layer and the number of filters, kernel, batch size, optimizer, and epoch values, the best model was identified with an accuracy of 99.47% and an F1-score of 99.44%. Additionally, our proposed best model achieved better results compared to previous studies and some transfer learning models, given the same optimizer, batch size, and epoch numbers.