search

uncbiag / ICON

A library for performing image registration using deep learning, regularized by inverse consistency
Other
42 stars 9 forks source link

Add multichanel support #54

Closed lintian-a closed 1 year ago

lintian-a commented 1 year ago

Support multichannel image as input to the network and similarity measure.

It is useful when there are multiple modalities such as MRI-T1, MRI-T2 and MRI-TI enhanced images.

This pull request contains the three following major changes:

  1. Expose an argument called input_channels for tallUNet2 function. This argument indicates the number of channels of one input image tensor.
  2. Refactor all the similarity measures to inherit from SimilarityBase class. And SimilarityBase class contains a variable member indicating whether this Similarity Measure requires inbounds_tag for evaluation.
  3. Add itk interface for multi-modality registration.
HastingsGreer commented 1 year ago

I really like the getattr trick to maintain backwards compatibility.

HastingsGreer commented 1 year ago

It looks good to me!