Skip to content

Latest commit

 

History

History

network

README.md

Model zoo

This page summarises the implemented network models in network.

All networks can be applied in 2D, 2.5D and 3D configurations and are reimplemented from their original presentation with their default parameters.

UNet

Reimplementation of

Çiçek, Ö., Abdulkadir, A., Lienkamp, S. S., Brox, T., and Ronneberger, O. (2016). 3D U-net: Learning dense volumetric segmentation from sparse annotation, In MICCAI 2016

Constraints
  • Image size - 4 should be divisible by 8
  • Label size should be more than 88
  • border is 44

VNet

Reimplementation of

Milletari, F., Navab, N., & Ahmadi, S. A. (2016). V-net: Fully convolutional neural networks for volumetric medical image segmentation, In 3DV 2016

Constraints
  • Image size should be divisible by 8

ScaleNet

Implementation of

Fidon, L., Li, W., Garcia-Peraza-Herrera, L.C., Ekanayake, J., Kitchen, N., Ourselin, S., Vercauteren, T. (2017). Scalable convolutional networks for brain tumour segmentation. In MICCAI 2017

Constraints
  • More than one modality should be used

HighRes3dNet

Implementation of

Li W, Wang G, Fidon L, Ourselin S, Cardoso M J, Vercauteren T, (2017). On the compactness, efficiency, and representation of 3D convolutional networks: Brain parcellation as a pretext task, In IPMI 2017

Constraints
  • Image size should be divisible by 8

DeepMedic

Reimplementation of

Kamnitsas, K., Ledig, C., Newcombe, V. F., Simpson, J. P., Kane, A. D., Menon, D. K., Rueckert, D., Glocker, B. (2017). Efficient multi-scale 3D CNN with fully connected CRF for accurate brain lesion segmentation, MedIA 36, 61-78

Constraints
  • The downsampling factor (d_factor) should be odd
  • Label size = [(image_size / d_ factor) - 16 ]*d_factor
  • Image size should be divisible by d_factor

Example of appropriate configuration: image_ size = 57, label_ size = 9, d_ factor = 3

To develop a new network architecture

  1. Create a niftynet/network/new_net.py inheriting BaseNet from niftynet.layer.base_net
  2. Implement layer_op() function using the building blocks in niftynet/layer/ or creating new layers
  3. Import niftynet.network.new_net to the NetFactory class in niftynet/__init__.py
  4. Train the network with python net_segment.py train -c /path/to/customised_config