Clinical diagnosis now plays a bigger part in today's healthcare system. Since brain cancer is the deadliest disease in the world, it is a significant concern in the field of medical imaging. Magnetic resonance imaging-based early and precise diagnosis may be beneficial for brain tumor evaluation and prognosis. In order for radiologists to employ computer-aided diagnostic procedures to assist them discover brain tumors, medical images need to be identified, segmented, and classed. There is an urgent need for an automated method since radiologists find the procedure of manually identifying brain tumors to be laborious and prone to mistakes. The method for precisely identifying and classifying brain tumors is thus introduced. There are five stages recommended for the procedure in terms of the tools and techniques used. To find the image's edges in the beginning, the original image is stretched with a linear contrast. The creation of a deep neural network architecture specifically designed for the goal of segmenting brain tumors occurs in the second stage. Finally, transfer learning is used to train a modified MobileNetV2 architecture for feature extraction. Finally, a controlled entropy-based method and a multiclass support vector machine (M-SVM) were used to choose the best features. Last but not least, M-SVM is used to classify images of meningioma, glioma, and pituitary tumors.

Biomedical image processing; brain tumor; deep learning; linear contrast stretching; segmentation.