Abstract

The deep learning architectures that have been used for brain tumor segmentation in the BraTS challenge have performed well for the WT, TC, and ET segmentations. However, these architectures generally have many parameters and require large storage capacity for the model. In this paper, we propose a Simple MyUnet3D to do segmentation on BraTS 2018 dataset. This proposed architecture was inspired by 2D U-Net and modified to do 3D image segmentation. Dataset divides into 2 parts, one part of training and the other for validation. From 285 data, 213 for training, and 72 for validating the model. The segmentation consists of 3 parts, whole tumor(WT), tumor core(TC), and enhanced tumor(ET). Even its simplicity, it produces a dice coefficient of 0.8269 at segmenting the whole tumor. Nevertheless, its performance in tumor core and enhanced tumor need to be developed. The simplicity and its result in segmenting the whole tumor have great potential to be better developed.

Original languageEnglish
Title of host publicationICICoS 2020 - Proceeding
Subtitle of host publication4th International Conference on Informatics and Computational Sciences
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Electronic)9781728195261
DOIs
Publication statusPublished - 10 Nov 2020
Event4th International Conference on Informatics and Computational Sciences, ICICoS 2020 - Semarang, Indonesia
Duration: 10 Nov 202011 Nov 2020

Publication series

NameICICoS 2020 - Proceeding: 4th International Conference on Informatics and Computational Sciences

Conference

Conference4th International Conference on Informatics and Computational Sciences, ICICoS 2020
Country/TerritoryIndonesia
CitySemarang
Period10/11/2011/11/20

Keywords

  • 2D U-Net
  • 3D Image Segmentation
  • BraTS 2018
  • BraTS Dataset
  • Brain Tumor Segmentation

Fingerprint

Dive into the research topics of 'Simple MyUnet3D for BraTS Segmentation'. Together they form a unique fingerprint.

Cite this