Adversarial Deep Structural Networks for Mammographic Mass Segmentation https://arxiv.org/abs/1612.05970 The overview is in isbi18_poster.pptx
This is the first work formulating the adversarial example to improve segmentation. Three aspects to understand adversarial deep structure model 1) optimal regularization for local smoothness 2) optimal data augmentation 3) force the model learn well on the boundary area.
crfcnn_combine.py is the executive file.
modelname = 'cnn' #'crfcomb'
modelname = 'cnn' #is for FCN model.
modelname = 'cnnat' #is for advesarial FCN model.
modelname = 'crf' #is for joint FCN-CRF model.
modelname = 'crfat' #is for adversarial FCN-CRF model.
modelname = 'cnncomb' #is for multi-FCN model.
modelname = 'cnncombat' #is for adversarial multi-FCN model.
modelname = 'crfcomb' #is for joint multi-FCN-CRF model.
modelname = 'crfcombat' #is for adversarual multi-FCN-CRF model.
crfcnn_combine_ddsm.py is the main file for ddsm.
.m files are for reproducing miccai 15 Deep Learning and Structured Prediction for the Segmentation of Mass in Mammograms.
The inbreast dataset can be downloaded from https://drive.google.com/a/uci.edu/file/d/0B-7-8LLwONIZM1djY2pRLWNUemc/view?usp=sharing
The DDSM-BCRP dataset can be downloaded from https://drive.google.com/a/uci.edu/file/d/0B-7-8LLwONIZU0l2N3hXdU96Y2M/view?usp=sharing
Please cite our paper as Zhu, Wentao, and Xiaohui Xie. "Adversarial deep structural networks for mammographic mass segmentation." arXiv preprint arXiv:1612.05970 (2016).
If you have any questions, please contact with me [email protected].
Supplement code and data in https://drive.google.com/file/d/0B5Hl9mO74DHvUEowa1hyWmVsMmc/view?usp=sharing . Maybe it is helpful for you to reproduce the results.