There are breaks in some areas of the segmentation results output by the network #4628
-
|
|
Beta Was this translation helpful? Give feedback.
Replies: 2 comments
-
|
the segmentation of these narrow tube structures is supposed to be hard and it's reasonable that the raw network output has these "breaks". I'm not sure what modality this is but I guess the intensity contrast on these structures is not apparent and the boundary is blurred. You can try some post-process techniques (CRF, level-set etc) to improve it). |
Beta Was this translation helpful? Give feedback.
-
|
Agree with @yilei-wu, it'll be difficult to provide concrete suggestions in this case. If there are assumptions about the connectivities, it might be interesting to incorporate these priors during the model training (eg learning based CRF). |
Beta Was this translation helpful? Give feedback.
the segmentation of these narrow tube structures is supposed to be hard and it's reasonable that the raw network output has these "breaks". I'm not sure what modality this is but I guess the intensity contrast on these structures is not apparent and the boundary is blurred. You can try some post-process techniques (CRF, level-set etc) to improve it).
Also I think this is really NOT the responsibility of the developers here to do the analysis for you or find you the reference paper ( especially when the given information is so limited)