Custom administering attention module for segmentation of magnetic resonance imaging of the brain

Abstract

Taking into account how brain tumors and gliomas are notorious forms of cancer, the medical field has found several methods to diagnose these diseases, with many algorithms that can segment out the cancer cells in the magnetic resonance imaging (MRI) scans of the brain. This paper has proposed a similar segmenting algorithm called a custom administering attention module. This solution uses a custom U-Net model along with a custom administering attention module that uses an attention mechanism to classify and segment the glioma cells using long-range dependency of the feature maps. The customizations lead to a reduction in code complexity and memory cost. The final model has been tested on the BraTS 2019 dataset and has been compared with other state-of-the-art methods for displaying how much better the proposed model has performed in the category of enhancing, non-enhancing and peritumoral gliomas.