An efficient automatic brain tumor classification using optimized hybrid deep neural network

Abstract

A significant topic of investigation in the area of medical imaging is brain tumor classification. Since precision is significant for classification, computer vision researchers have developed several approaches, but they still struggle with poor accuracy. In this paper, an automatic optimized hybrid deep neural network (OHDNN) is suggested for brain tumors. The proposed approach consists of two phases such as pre-processing and brain tumor classification. At first, the images are composed of the data, and then the collected imageries are pre-processed by using the following steps such as image enhancement and noise removal. Then the pre-processed images are fed to the classification stage. For the classification process, in this paper, OHDNN is used. The HDNN is a combination of a convolution neural network and long short-term memory (CNN-LSTM). Here, the CNN classifier is used for feature map generation and the classification process LSTM classifier is used. Besides, to improve the performance of the CNN-LSTM classifier, the parameter extant in the classifiers is randomly selected utilizing the adaptive rider optimization (ARO) algorithm. For the experimental process, an MRI image dataset is utilized. The experimental results show proposed approach attained the maximum accuracy of 97.5.

Keywords

• Brain tumor classification
• Convolution neural network
• Long short-term memory
• Rider optimization
• CNN-LSTM