On comparing optimizer of UNet-VGG16 architecture for brain tumor image segmentation

Anindya Apriliyanti Pravitasari, Nur Iriawan, Ulfa Siti Nuraini, Dwilaksana Abdullah Rasyid

Research output: Chapter in Book/Report/Conference proceedingChapterpeer-review


Image segmentation as computer-based diagnostic systems in brain tumor magnetic resonance imaging (MRI) plays important role in supporting the medical diagnosis. The clear shape of tumor visualization is the main objective in image segmentation. Several methods have been developed in the deep learning area under the U-Net architecture. The UNet-VGG16 with transfer learning + dropout is a new architecture that hybrids the U-Net with VGG-16 added by the transfer learning + dropout regularization. The dropout scenario is a way to minimize the overfitting effect as the nature of VGG-16 contains different nonlinear hidden layers and complex relationships that may result in overfitting. This new architecture is proved to be fast and accurate in segmenting the MRI of low-grade gliomas in fluid-attenuated inversion recovery sequence. However, the previous model is yet to concern the effect of the different optimizers on the overfitting issue. Therefore, this study aims to investigate the best optimizer for UNet-VGG16 with transfer learning + dropout that could deal with overfitting. This study had succeeded to recognize Adamax as the best optimizer.

Original languageEnglish
Title of host publicationBrain Tumor MRI Image Segmentation Using Deep Learning Techniques
Number of pages19
ISBN (Electronic)9780323911719
ISBN (Print)9780323983952
Publication statusPublished - 1 Jan 2021


  • Adamax
  • Image segmentation
  • Optimizer
  • Overfitting
  • U-Net
  • VGG-16
  • low-grade gliomas


Dive into the research topics of 'On comparing optimizer of UNet-VGG16 architecture for brain tumor image segmentation'. Together they form a unique fingerprint.

Cite this