Large sets of medical images in the hospital's image archiving and communication system combined with advanced high-performance computing enable the acceleration of machine learning techniques for recognizing clinical findings, images and accurate diagnosis of specific diseases. With a large enough amount of data and adequate training, deep-learning models can achieve high accuracy. Medical images have massive potential for deep learning applications, and high-performance models can be used for practical purposes. Accordingly, this research was conducted on a publicly available data set from the glioblastoma archive of the University of Pennsylvania Medical Health System. The data archive consists of brain magnetic resonance studies performed in T1, T2 and T2-flair sequences on 650 patients. From the entire dataset, 130 images of identical brain slices were exported and used for manual glioblastoma labelling and further application in a deep learning model for three different architectures. According to the results and comparing all three architectures, it is concluded that image classification and object detection have many false positive detections. The image segmentation model with the same parameters as the other two models could have better results on the test data. However, by adjusting the relevant parameters, the image segmentation model achieves significant results better than the other two methods. According to the conclusions of many studies, U-Net will be crucial for future development due to its applicability in a 3D environment, which enables the development of industrially applicable algorithms.