I am training a U-Net image segmentation on whole CT & MRI images. how can I handle un-annotated areas? I am working with huge tissues and it’s impossible to annotate all or the vast majority of the tissue, Aim annotated selected tissue structures of interest to me. That means that in many polygon (mesh) I’m generating, there is a segment that’s not annotated.
Would it affect the U-Net negatively by indirectly indicating that the un-annotated area is negative to one category or another, although it’s not negative? How do I handle this important case? Does it make sense to mask the image to only the annotated parts, such that un-annotated regions are black?
You may train 2D network with available data and use it to annotate the rest of the data. Then, you can train the 3D network. Of course, this may need correction.
Otherwise, as you mentioned you may set these slices to zero or t random values. However, if these unannotated slices are too much, this may not gi e you the best results.
- Train network using the annotated 2D images
- Predict the annotation of the unannotated images using the trained network
- Use the annotated images and the predicted images to train a new network using 3D volumes
- Badges
- Science topic
- Similar topics
- Biological Science
- Anatomy
- Tissue