A Classification Method for Brain MRI via AlexNet

Abstract

The number of people dying from brain tumors is increasing day by day. Early diagnosis is very important in the treatment planning and evaluation of the treatment outcome of brain tumors. A patient with a brain tumor may be more likely to survive by applying the right treatment methods if the disease is diagnosed early. Medical imaging methods have an important role in the identification and diagnosis of brain tumors. One of the most popular medical imaging methods is Magnetic Resonance Imaging, MRI. Determining the presence of tumors and tumor characteristics from MRI is done by specialists. In today's technology, computer-assisted detection applications make great contributions to the field of medicine. Computer-Assisted Detection (CAD) software helps radiologists to detect abnormalities in medical images by using advanced pattern recognition and image processing methods. This software not only saves time for radiologists but also minimizes possible errors in the decision-making phase. In this study, deep features were extracted from a total of 942 MRIs with 599 tumor and 343 normal class labels using the AleXNet-based deep learning model, and classification was performed with the K Nearest Neighbor Classifier (KNN) algorithms. In this study, 1000 deep features were extracted from the MRI data with the trained weights of the fully connected layer named “fc8” of the AlexNet model. Then, these features were reduced by Relieff feature selection algorithm, and the performance of the proposed method was increased. A weighted KNN classifier was used in the classification phase. With the proposed method, 87% classification accuracy was achieved.