Learning Volumetric Segmentation for Lung Tumor



Demo

Abstract

Convolutional neural networks have recently been applied to various computer vision as well as medical image analysis tasks. Despite their popularity, most of the existing methods are tackling problem where data is presented in the form of 2D images (e.g., semantic segmentation). However, in medical image analysis scenario, data often contains 3D (i.e., volumetric) contents. In this paper, we present an approach that addresses the challenging 3D medical image segmentation. Our model is end-to-end trainable on CT images and learns to predict volumetric segmentation outputs. To handle the imbalance data distribution between the foreground and background region, we employ the dice loss and the focal loss during optimization. As such, our model can precisely segment out sparse objects within dense region. Extensive experimental evaluations on a challenging benchmark demonstrate that out model performs favorably against the state-of-the-art methods.
Paper
Coming Soon

ICIP 2018
Presentation Slides

Citation

Jhih-Yuan Lin, Min-Sheng Wu, Yu-Cheng Chang, Yun-Chun Chen, Chao-Te Chou, Chun-Ting Wu, and Winston H. Hsu, "Learning Volumetric Segmentation for Lung Tumor", 2018.


BibTex
        @inproceedings{VolSegNet,
          author    = {Lin, Jhih-Yuan and Wu, Min-Sheng and Chang, Yu-Cheng and Chen, Yun-Chun and Chou, Chao-Te and Wu, Chun-Ting and Hsu, Winston H.}, 
          title     = {Learning Volumetric Segmentation for Lung Tumor},
          year      = {2018}
        }
      
Code
Coming Soon

The Proposed Accurate, Efficient, and Compact Network
Coming Soon

Results
Coming Soon