Abstract
Automatic classification of brain tumor types is essential to accelerate the treatment process and increase the patient’s survival rate. In today’s world, magnetic resonance imaging is used for determining the brain tumor types in an effective manner. The manual classification of brain tumor consumes more time and also applicable for a limited number of images, so a deep learning-based model is proposed in this research paper for automatic classification of brain tumor types. At first, the brain images are collected from T1-weighted contrast-enhanced magnetic resonance image dataset. Then, the normalization technique and histogram of oriented gradients are employed to improve the visible level of the collected raw brain images and to extract the feature vectors from the normalized brain images. The histogram descriptor effectively describes the edge and contour features of the images related to other feature descriptors. Further, the extracted features are given as the input to convolutional neural networks to classify meningiomas, gliomas, and pituitary tumors. The hard swish-based RELU activation function is included in convolutional neural networks that effectively improve the classification performance and learning speed. In the experimental phase, the proposed model achieved 98.6% of accuracy that is better compared to the existing algorithms like deep convolutional neural network with transfer learning and fine-tuning.