Classification of CT/MR Images using Convolution Neural Network

Abstract

Brain tumor has been the most hazardous disease all around the world. Brain tumor is a collection of tissue that has been formed through the gradual accumulation of irregular cells. It originates when cells in the brain produce irregular formations. A Convolutional Neural Network (CNN) technique for classification of Magnetic Resonance (MR) images are proposed in this paper. This study has used 3264 brain images dataset containing Glioma tumor (926 images), Pituitary tumor (901 images), Meningioma tumor (937 images) and have no-tumor (500 images). In this work 80% of the training images which are 2296 MR brain images are taken for the training of propose CNN model and 20% of the training images which are 576 MR brain images are taken for the validation of propose CNN model. And 392 MR brain images are used for testing of propose CNN model. In the preprocessing phase fuzzy based method is introduced for enhancement of brain images dataset. The propose CNN model with categorical cross-entropy loss function as well as parameter adjustment give us 97.60% accuracy for classification of brain tumor into glioma, meningioma, no-tumor, and pituitary tumor. Finally, this work will help to make an effort to explore the future in the field of Artificial Intelligence (AI). It also includes several research tracks that may be explored in the future and will be useful in the field of medical science.