U-Net for Efficient Liver Segmentation from 3D CT Data

EasyChair Preprint 54

Abstract

In order to diagnosis the liver cancer in early stage, the liver segmentation technique for image processing had been challenged by the non-homogeneous Hounsfield Unit (HU) of nearby area around the liver in 3D computed tomography image (CT image) and the similar morphological shape of nearby organ. The previous study reported the high accuracy and precision segmentations by using hybrid method, (combining image-based and model-based method). In order to reduce the processing time of segmentation, this U-Net model was developed by using the training dataset (16 images) and testing dataset (4 images) from the Segmentation of the Liver 2007 (SLIVER07). The results showed the high accuracy segmentation operated under 0.02 seconds per slice.

Keyphrases: 3D CT Data, U-Net, liver segmentation

@booklet{EasyChair:54,
  author    = {Boonnatee Sakboonyarat and Pinyo Taeprasartsit},
  title     = {U-Net for Efficient Liver Segmentation from 3D CT Data},
  doi       = {10.29007/sfg1},
  howpublished = {EasyChair Preprint 54},
  year      = {EasyChair, 2018}}