Error Corrective Boosting_Jully.2017

This paper can be considered as the former version of the QuickNAT.

Summary

Problem:

1. Labeled image data is very limited.

Solution:

1. Automatically create auxiliary labels on initially unlabeled data with FreeSurfer and use them for pre-training.
2. Introduce ECB (Error Corrective boosting) for fine-tuning the network with manually labeled data.
3. Introduce SkepDeconv-Net.
4. Segment a 3D scan in 7 secs
5. Outperforms the latest state-of-the-art F-CNN models.