Brain Tumor Image Classification Based on Wide Dense Convolution Neural Network

Brain tumor diagnosis requires accuracy and speed to promptly provide treatment to patients. Currently, specialist doctors diagnose brain tumors using MRI images. Therefore, research on brain tumor classification from MRI images has been developed to increase the accuracy of tumor classification and reduce processing time. This research presents a convolutional neural network model for classifying brain tumors from MRI images. It proposes a wide-density block model with VGG12. The objective is to improve the accuracy of brain tumor classification from MRI images, reduce the computation time, and reduce the number of model parameters. The proposed model improves the Dense Block based on DenseNet121 by increasing the Conv2D layers from 2 to 3 and bypassing the image feature maps between layers in a wide plane. The improved Wide Dense Block is combined with the VGG16 model, which has fewer layers to 12. The experiments are conducted on a brain tumor image dataset from www.kaggle.com, which contains three classes of brain tumor images and one class of normal brain images, totaling 7,023 images. The proposed model achieves the highest accuracy of 99.87 % when compared to ResNet50, VGG16, InceptionV3, DenseNet121, and MobileNetV2 models, which achieve accuracies of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$77.07 \%, 90.74 \%, 97.34 \%, 96.63 \%$</tex>, and 99.55 %, respectively. The proposed model has 2,521,606 parameters, which reduces its processing time.