Recent advances in the diagnosis of brain tumors have played an important role in improving the condition of patients and increasing the life expectancy of patients. Deep learning techniques have been considered in the area of tumor diagnosis with great accuracy, and therefore has been expanded rapidly in this area. Diagnosis of a brain tumor with the help of the computer includes tumor¬ detection, ¬segmentation and classification. In this study the brain tumor diagnosis system using MRI has been considered and deep learning approaches have been considered to diagnose brain tumors. Also, differences and similarities of many techniques in terms of their performance in the three processes of brain tumor diagnosis are presented. One of the most popular networks used in deep learning is the convolutional neural network. In this thesis, using an improved U-Net convolutional neural network with more concatenation has been proposed done in order to extract better features and better model training. The accuracy of brain tumor segmentation using the improved U-Net network is 80.48%. Also, using the improved U-Net network Encoder section, the brain tumor classification was performed with 97% accuracy. To better eva‎luate the results of classification operations on test data, the K-Fold cross-validation method has been used, which has an average network accuracy of 7-Fold, 95.64. Combining the improved U-Net network and the optimal classification network, a network is provided for both operations. Brain tumor images are used as network input, and at the network output, classification and segmentation operations are performed simultaneously.

محمد رضا احمدزاده

نادر كريمي، مهران صفاياني