Deep Learning Methods for Brain Tumor Segmentation

MRI, or magnetic resonance imaging, is one of the most recent medical imaging techniques. It allows one to visualize organs and soft tissues in different planes of space with great precision. A single person's brain is scanned using MRI in several slices through a 3D anatomical viewpoint. However, it is difficult and time-consuming to manually segment brain tumors from MRI images. Furthermore, automatic segmentation of brain tumors using these images is noninvasive, avoiding biopsy and improving the safety of the diagnosis procedure. This chapter enriches the body of knowledge in the field of neuroscience. It describes a highly automated technique for segmenting brain tumors in multimodal MRI based on deep neural networks. An experimental study was carried out using the Brain Tumor Segmentation (BraTS 2020) dataset as a proof of concept. The accuracy, precision, sensitivity, and specificity exceed 99.3%. In addition, the achieved intersection over union and loss are 85.69% and 0.0177. The obtained results based on the proposed method are validated by comparing them to real values found in the state of the art.