Lung Tumor Segmentation by Fusing 2D and 3D Models

EasyChair Preprint 3185

Abstract

In this study, an automatic lung tumor segmentation model is proposed. Over the past decade, deep neural networks have shown many strengths in the field of image segmentation. Especially in the field of medical imaging, many network models have been developed in recent years. These networks mostly use data for training as 2-dimensional (2D) data because they do not require a lot of hardware resources and are easily visualizing. However it is not fully utilized 3-dimensional (3D) structure, so there are also number of studies using 3D model for training. Our proposed method uses a model that fusion both 2D and 3D data for training. The results of proposed model are evaluated on the data set of task 4th of the StructSeg competition with very positive results, it opens up many later research directions in medical image processing.

Keyphrases: Lung Tumor Segmentation, Lung tumor, U-Net, image segmentation, residual network, structseg competition

@booklet{EasyChair:3185,
  author    = {Minh-Trieu Tran and Soo-Hyung Kim and Hyung-Jeong Yang and Guee-Sang Lee},
  title     = {Lung Tumor Segmentation by Fusing 2D and 3D Models},
  howpublished = {EasyChair Preprint 3185},
  year      = {EasyChair, 2020}}