Performance Optimization of Brain Tumor Detection and Classification Based MRI by Using Batch Normalization Algorithms in Deep Convolution Neural Network

Abstract

Brain tumor is represented as an essential part of critical cancers around the world. These cells multiply and accumulate uncontrolled, forming a mass or lump that can interfere with normal brain function. Primer detection systems not only took too must time in analyzing and setting error, but also extended more datasets to become overfitting, more computation time, and lack accuracy. Supervised ML and traditional CNN are not convenient for estimating the vita feature engineering in larger datasets and they need to be modified using normalization techniques in deep convolutional Neural Networks (CNNs) architectures. The proposed of the research MRI image datasets were evaluated and combined with two popular benchmark data sets, Kaggle, and BRATS. This main objective is to reduce the computational cost avoid overfitting and underfitting and then improve the classification accuracy. In addition, this paper follows the concept of the CNN model and evaluates the modified DCNN with six normalization layers benefits acceptable results with batch normalization techniques and the average number of epochs in a limited time. In this regard, we exploited to extend inside the layer DCNN for the problem of brain tumor classification. This model achieved the best result for the enhanced dataset, with a training accuracy of 99.9%, 98.9% in validation accuracy, 0.0074 in training loss, and a validation loss of 0.0566 in validation loss.