Data Augmentation by Adaptative Targeted Zoom for MRI Brain Tumor Segmentation

This study presents a novel data augmentation methodology to enhance the scientific outcomes achieved during the Brats 2023 challenge. The 3D-UNet neural network, initially proposed by Ronneberger et al. in 2015 for biomedical image segmentation, is described. Its performance is evaluated in the segmentation of human brain tumors utilizing authentic MRI data from the BraTS 2023 challenge. The specifics of the data augmentation algorithm and its importance in the context of MRI images of this nature, as well as the utilized network architecture, are briefly expounded. Finally, future directions are outlined.