Replies: 1 comment 3 replies
-
Hi @kirrog, It is great to hear you're using MONAI Label for this type of images. Is it for teeth and mandible segmentation? I'm working with @dgmato on the same task. would you like to join forces? We have around 40 volumes fully segmented for this. For this task we're using the multistage approach developed for vertebra segmentation: https://github.com/Project-MONAI/MONAILabel/tree/main/sample-apps/radiology#multistage-vertebra-segmentation With regards to your questions:
For a multilabel segmentation task, MONAI Label needs one file containing one label map with all classes/labels.
Yes. MONAI Label only considers the labels defined in the label dictionary. A similar use case is here: https://github.com/Project-MONAI/MONAILabel/blob/main/sample-apps/radiology/lib/configs/deepedit.py#L42-L51 Hope this helps, |
Beta Was this translation helpful? Give feedback.
-
Hi everyone!
We are trying to learn deepedit, using monai radilogy app as template, to segment many labels, about 20.
We've changed
self.labels in configs/deepedit to vocabulary contains pairs of label_name:label_number, similary to original
And place in root of "studies" directory all CBCT as nifti
In root/labels/final we've placed nifti in which we place segmentation all labels, as label_num in voxels where segmentation should be
We try to learn on all segments and found out, what if one label is huge, and other is tiny, deepedit prefer to segment the huge one. It's logical because of loss. So we tried to segment tiny ones by one, but monai used all segmentation map as segmentation of this label.
We tried to segment many labels, but not all, and when got results near what did we expect, but now we fear of probability, what monai may do some missunderstanding of our data, and so we have worser results, when we may, because of this missunderstangind
Can you please explain:
Beta Was this translation helpful? Give feedback.
All reactions