paper_outline_pu_learning_brain_tumor_segmentation

[xuyanwu]

No description provided.

[xuyanwu]

write the outlines in the following format:

Introduction

• introduce brain tumor segmentation task with the different supervised setting, handcrafted setting, weak supervision, supervised settings.
• introduce the difficulty of brain tumor segmentation. (collecting enough label data)
• introduce the recent model for segmenting brain tumor (true label traning).
• the reason why we apply weak-supervised brain tumor segmentation with box label and treat it as PU learning problem. (motivation of our method)
  • the labor comsuming of labeling 3-D brain tumor slices.
  • introduce the PU learning and the relationship between PU learning and weak-supervised segmentation.
• the segmentated example support our method.
• contribution paragraph:
  • We propose a novel method for weak-supervised brain tumor segmentation
  • We suggest apply PU-leanring method on this box labeled weak supervised segmentation task

Related Work

Method

Notation

Proposed method

• network structure for segmentation:
  • 3-D U-net segmentation introduction
• PU-boxed weak-supervised learning
  • the whole idea of how to generate the box labeled segmentation data and how to extract patch for training
  • the graph of our proposed model
  • formulation of the plain box segmentation model
  • formulation of PU-segmentation model

Experiments

• BraTS data
  • training settings, data introduction
  • segmentation output: gt label training vs plain box segmentation vs PU box segmentation
  • dice score, Hausdorff distance: gt label training vs plain box segmentation vs PU box segmentation

conclusion

[mgong2]

looks like a reasonable outline. But do we need to have the background section?