3D Brain MRI Segmentation using Deep Neural Network

Abstract

Magnetic Resonance Images (MRI) have been utilized by radiation oncologists to identify the size of tumours within the brain.The accurate identification of brain tumours from 3D MRI images is essential for proper diagnosis. Our proposed model introduces a network capable of segmenting 3D images using sparsely labeled data. This network is an enhanced version of the u-net architecture with attention network and utilizes 3D operations, without the need for a pre-trained network. The effectiveness of this approach was evaluated on the well-known BraTS 2018 brain dataset and achieved a Dice Similarity Coefficient score of (0.95, 0.89, 0.95) for the complete, core, and enhancing regions, respectively