Automatic Brain Tumor Segmentation from MRI Scans using U-net Deep Learning

Abstract

Segmenting brain tumors is crucial for improving diagnosis, prognosis, and therapy options. Brain tumors treatment must begin right away, because identifying brain tumors manually is difficult and time-consuming. It might be feasible to better accurately identify brain tumors and improve treatment planning by using automatic segmentation algorithms. Convolutional Neural Networks (CNN) have recently been shown to be effective for the task of segmenting medical images. In the paper, a deep learning-based approach for segmenting brain tumors from MRI datasets is proposed. This technique could help radiologists identify brain tumors. The suggested approach focuses on combining various processes; it begins with a preprocessing phase to enhance the appearance of tumors in various MRI maps. To enhance the contrast and intensity homogeneity of brain tumors, the Edge Enhancing Diffusion filtering (EED) filtering method is utilized, which is regarded as a sort of hard attention because it emphasizes the brain tumors regions. The redesigned decoder path used in the proposed U-net based network architecture employs the attention module. In order to improve and capture long-range relationships and give the network the ability to distinguish s from background noise, channel attention mechanisms as well as spatial attention mechanisms are used. The BRATS 2020 challenge dataset is used to assess the suggested approach. Average DSC scores were 90.80% and 92 % on the sensitivity scale.