Brain Tumor Detection and Classification Using Deep Learning From Magnetic Resonance Images

Abstract

Brain tumors are the most conventional and aggressive illness and have an extremely low life expectancy in its most severe form. Uncontrolled growth as well as proliferation of skull cells results in the formation of brain tumors. Given the difficulties of tumor biopsies, deep learning-based brain tumor analysis frequently makes use of three dimensional (3D) magnetic resonance imaging (MRI). In this article, a detailed analysis of different deep learning method is presented for the classification of multiclass brain tumor which may increase the level as well as efficiency of MRI engines in detecting the disease. Before using the image dataset to stabilize the output of four previously trained convolutional neural network (CNN) models, it undergoes a variety of data pre-processing techniques. The pre-trained prototypes used in this study are EfficientNet, Xception, MobileNet, and ResNet-50. Accuracy is shown to rise as the period progresses where the highest accuracy of EfficientNet, Xception, and MobileNet is 98.18%, 97.23%, and 96.96%, respectively where ResNet-50 comes with a moderately much lower accuracy score of 95.66%. The supremacy of EfficientNet over all other models is discovered in Precision, Recall as well as F1-Score while classifying brain tumor cells.