| Abstract [eng] |
State-of-the-art convolutional neural network architectures and their application to brain tumor segmentation of MRI images are analyzed. We identify models that over the last few years have achieved the best results in semantic segmentation tasks in terms of Sorensen-Dice coefficient (F1). We also experiment with 3 neural network architectures (U-Net, Attention U-Net and SegNet) in order to improve results of the segmentation of low grade gliomas in the axial plane. Segmentation results of SegNet are improved by modifying the network with recurrent residual blocks. |
