Download PDFOpen PDF in browser

Glioblastomas Brain Tumor Segmentation Using Optimized Three-Dimensional (3DU-Net) Model

EasyChair Preprint 10113

10 pagesDate: May 12, 2023

Abstract

Automated segmentation  is a computerized technique that helps to find tumor location, size, and shape. Human   segmentation is error prone, time consuming, and needs an expert radiologist. In our study, we developed a customized 3D U-Net model that processes 3D volumetric images for multiclass tumor segmentation.This framework is modified in such a way that  the gradient flow is better for finding accurate output. The BraTS 2020 dataset is used to train  this  network with  end-to-end  learning strategy followed by defining the proper skip connection from encoder  to decoder.  In model  evaluation, binary cross-entropy with Dice loss functions is utilized.Testing samples are predicted and classified into  three  regions: whole tumor (WT), tumor core (TC),  and enhancing  tumor (ET). Model performance is evaluated through  Dice  coefficient metrics for each class. On the basis of this model, experiments were carried out on the BraTS 2020 dataset which could be considered as a validated benchmark. The  segmentation’s obtained results  have  been validated  with  ground  truth  references by computing the Dice Metric parameter.Our clinical partners  have   attested that  the   proposed  tool   could achieve great performances. The aim of this research is to make  an  advanced  tool which  could  help  radiologists  to make  a  more  accurate  diagnosis.  It could  also assist clinicians in the early detection of brain tumors.

Keyphrases: 3D U-net, Brats dataset, Deep Learning (DL), Gliomas, Resonance Imaging (MRI), Segmentation

BibTeX entry
BibTeX does not have the right entry for preprints. This is a hack for producing the correct reference:
@booklet{EasyChair:10113,
  author    = {Amor Feriel and Mzoughi Hiba and Njeh Ines and Benslima Mohamed},
  title     = {Glioblastomas Brain Tumor Segmentation Using Optimized Three-Dimensional (3DU-Net) Model},
  howpublished = {EasyChair Preprint 10113},
  year      = {EasyChair, 2023}}
Download PDFOpen PDF in browser