Download Source Code
This Caffe branch contains the implementation of the topology aware loss function
proposed in  for learning toplogically-plausible segmentations using fully convolutional networks (FCN).
The proposed loss encodes smoothness of, and topological constraints between, segmented regions of
spatially-recurring, multi-part objects (e.g. several glands, each with
lumen and epithelium) into the learning of FCNs.
/!\ A tensorflow version is now available at the following URL.
The code and models released here are provided "as is" without any implied warranties for any particular purpose. In no event shall the authors of this site be liable for any damage or tort arising from the use of this source code, even if we are advised of the possibility of such damage. This source code is not for commercial use. For commercial interest please contact Prof. Ghassan Hamarneh.
If you use this code, please cite the following paper:
Topology Aware Fully Convolutional Networks For Histology Gland Segmentation
Aicha BenTaieb and Ghassan Hamarneh
Lecture Notes in Computer Science, Medical Image Computing and Computer-Assisted Intervention (MICCAI),
volume 9900, pages 460-468, 2016
© SFU Medical Image Analysis Lab, 2016