Can we generate a single multi-structure segmentation network from partially annotated datasets?
In this conference paper Sierra’s researchers in conjunction with Brigham and Women’s Hospital answer such question, showing that a) it is equivalent to use a single unet and b) that adding convolutions on the skip connections further improve segmentation results.
The full article can be found here: https://link.springer.com/chapter/10.1007%2F978-3-030-00946-5_22
