Brain Tumor Classification Using Convolutional Neural Network

Abstract

Deep learning (DL) network is prioritized for accuracy at higher levels and inherent automated feature extraction along with large amount of labeled data and computing power. In medical application, deep learning is used to detect cancer cells automatically. Deep learning consists of more than 150 hidden layers whereas neural network consists 3 hidden layers. Deep Convolutional Neural Network (DCNN) has achieved a great success in computer vision. Stirring by the structure of visual cortex, CNNs embedded with multiple hidden convolutional layers between the input and output layers have a capability of extracting higher level representative features and non-linear properties. Combination of DL with CNN has excellent results on medical field including classification of skin cancer, diabetic retinopathy detection and brain timer segmentation. DCNN is used to classify images of image net with help of an Alexnet model. By convolving small filters with the input patterns, features extraction is done followed by selection of the most distinguishing features and then start to train the classification network. In this paper, different classifiers such as Alex net, Google Net and ResNet are chosen for their error rate architectures. The malignant tumor images that occur in the spinal cord and brain at different stages are the input to the classification system. The main features extracted during pre-processing fed as input to the classifier networks which has been already trained. To evaluate the extracted characteristics, the performances of the classifiers are validated based on architecture, repetition, time consumption and accuracy with respect to number of iterations.