<![CDATA[Brain tumor is a deadly disease where 3.7% per 100,000 patients have malignant tumors. To analyze brain tumors can be done through magnetic resonance imaging (MRI) image segmentation. Automatic image analysis process is needed to save time and improve accuracy of doctor diagnoses. Automatic segmentation can be done with deep learning. U-NET is one of the methods used to segment medical images because it works at pixel level. By applying the ReLU and Adam Optimizer activation function, this method can solve the problem of segmenting brain tumors. Dataset for the training and validation process using BRATS 2017. Several hyperparameters are applied to this method: learning rate (lr) = 0.0001, batch size (bz) = 5, epoch = 80 and beta (b_1) = 0.9. From a series of processes carried out, accuracy of the U-NET method is calculated by Dice Coefficient formula and results in following accuracy values, during training of 90.22% (Full Tumor), 78.09% (Core Tumor) dan 80.20% (Enhancing Tumor).]]>
Copyrights © 2020
