TransUNet with Attention Mechanism for Brain Tumor Segmentation on MR Images
Enhao Wang, Yaqi Hu, Xiaoniu Yang, Xiaolin Tian
Abstract
Malignant brain tumor is a serious disease to human, causing cancer even death. Surgery is the main method for treatment of brain tumor. It’s a key issue to detect the existence and position of brain tumor, for doctors to analyze the brain image and make a plan for treatment. This task, brain tumor segmentation, is mainly performed on MRI (Magnetic Resonance Image). In recent years, machine learning and deep learning received extensive attention and developed rapidly. This passage introduces a new deep learning model, TransNUNet, to perform tumor segmentation task on brain MRI dataset. This model introduces attention mechanism, Cbam, to TransUNet, and refines loss function. Research shows that TransNUNet has a higher Dice score than other two image segmentation nets, U-Net and TransUNet, which proves its potential on this task.