An Automatic Brain Tumor Segmentation Using 3D Residual U-Net

• Main Authors: Hardani, Dian Nova Kusuma, Nugroho, Hanung Adi, Ardiyanto, Igi
• Format: Conference or Workshop Item PeerReviewed
• Language: English
• Published: 2022
• Subjects: Electrical and Electronic Engineering
• Online Access: https://repository.ugm.ac.id/279364/1/Hardani_TK.pdf; https://repository.ugm.ac.id/279364/; https://ieeexplore.ieee.org/document/10057670
• Institution: Universitas Gadjah Mada

A brain tumor is a group of abnormal neuronal cells that can spread and modify brain structure. Brain tumors are one of the deadliest diseases ever identified. Appropriate diagnostic and surgical planning for brain tumor patients increases survival rates and treatment options. Precise brain tumor segmentation determines surgical site and diagnosis. However, proper segmentation of brain tumors is difficult due to the diverse forms and appearances of brain tumors. This study provides a method for segmenting sub-areas of brain tumors using a ResU-Net model. The proposed model is effective combines encoding residual blocks using attribution mapping the U-Net model’s component to enhance the procedure for learning. It is meant to improve the comprehensive training method and resolve the gradients issue. Using the BraTS 2020 benchmark dataset, the proposed model was assessed. The results proved the superiority of the proposed technique, with whole tumor, tumor core, and enhancing tumor earning dice scores of 0.914, 0.903, and 0.882, respectively.