Brain Tumor Classification Using Convolutional Neural Networks and Deep Learning

Image classification represents a widely researched current topic, with hundreds of applications in medical imaging. The classification of brain tumors cannot be an exception, it is an applications where various machine learning techniques have been tried out already and convolutional neural network (CNN) architectures lead the benchmark records. Several brain MRI datasets were involved in the search of the best performing CNN, and these studies found better and better solutions for a specific set of data. In this paper we propose a CNN architecture which is trained and tested on three different datasets, including classification problems of two, three, and four classes, with the only architectural difference of using one neuron for each class in the output layer of the fully connected network part. The proposed network produced classifications of fine accuracy, with misclassification rates of 1.1 %, 2 %, and 4 %, when deployed in the two-class, three-class, and four-class classification problems, respectively. These benchmarks indicate that the proposed network can compete with state-of-the-art solutions. There are still several open directions for further optimization of the decision making system.