• Back to README

• Weakly Supervised Segmentation

  • (**) DeepCut: Object Segmentation From Bounding Box Annotations Using Convolutional Neural Networks, TMI, 2017

Weakly Supervised Segmentation

(**) DeepCut: Object Segmentation From Bounding Box Annotations Using Convolutional Neural Networks, TMI, 2017

Problem

• many modern medical image analysis methods that are based on machine learning rely on large amounts of annotations to properly cover the variability in the data.
• However, the effort for a single rater to annotate a large training set is often not feasible.

Contribution

• We extend the basic idea with recent advances in CNN modelling and propose DeepCut, a method to recover semantic segmentations given a database of images with corresponding bounding boxes.

Results

Limitations

• The default CRF parameters θ and ω in [19] (see Tab. I) were not appropriate for medical images and required tuning.
• Further, the time required for training of one epoch was approximately 9 minutes and inference during testing (including CRF) was less than 8 minutes for the largest MR volume.
• these might be modality and problem dependent, requiring adjustment when translating DeepCut to new segmentation problems.