Curriculum semi-supervised segmentation
Authors:
- Hoel KervadecHKHoel Kervadec
- Jose DolzJDJose Dolz
- Eric GrangerEGEric Granger
- Ismail Ben AyedIAIsmail Ben Ayed
Preprint
Abstract
This study investigates a curriculum-style strategy for semi-supervised CNN segmentation, which devises a regression network to learn image-level information such as the size of a target region. These regressions are used to effectively regularize the segmentation network, constraining softmax predictions of the unlabeled images to match the inferred label distributions. Our framework is based on inequality constraints that tolerate uncertainties with inferred knowledge, e.g., regressed region size, and can be employed for a large variety of region attributes. We evaluated our proposed strategy for left ventricle segmentation in magnetic resonance images (MRI), and compared it to standard proposal-based semi-supervision strategies. Our strategy leverages unlabeled data in more efficiently, and achieves very competitive results, approaching the performance of full-supervision.Keywords
All Related Versions
- Version 1, 2019-04-10, ArXiv
- Version 2, 2019-07-26, ArXiv
- Published version: , 568.
This publication has 19 references indexed in Scilit:
- Constrained-CNN losses for weakly supervised segmentationMedical Image Analysis, 2019
- On Regularized Losses for Weakly-supervised CNN Segmentation2018
- Unsupervised Domain Adaptation for Automatic Estimation of Cardiothoracic Ratio2018
- ASDNet: Attention Based Semi-supervised Deep Networks for Medical Image Segmentation2018
- Semi-supervised Learning for Segmentation Under Semantic Constraint2018
- A survey on deep learning in medical image analysisMedical Image Analysis, 2017
- Semi-supervised Learning for Network-Based Cardiac MR Image Segmentation2017
- Semi-supervised Deep Learning for Fully Convolutional Networks2017
- Deep Adversarial Networks for Biomedical Image Segmentation Utilizing Unannotated Images2017
- Semi-supervised Segmentation of Optic Cup in Retinal Fundus Images Using Variational Autoencoder2017