I’m trying to perform inference using the pretrained model from the 3 stage vertebrae segmentation pipeline from MONAI Label, but have been running into a few issues on a Windows machine with Slicer version 5.2.1:
I’ve been able to start the server and run the first step localization_spine without runtime errors, but unfortunately it produces empty segments. I’ve tried this using a nifti file from both Verse2019 and Verse2020, which are the datasets the 3 stage pipeline were originally trained on. I’ve tried other models such as the segmentation_spleen and the segmentation_full_ct from here:radiology_full_ct-upgraded-HYBRID - Google Drive, and those have produced mostly accurate output labels.
I also tried running the entire vertebrae_pipeline, but this resulted in the following error:
It is good that you managed to run the vertebra model on the Medical Segmentation Decathlon dataset.
The vertebra segmentation model was trained only on a portion of the VerSe dataset (~60 volumes). I’ve also had similar issues with that dataset and I think it was because some files were corrupted or had wrong NIfTI headers.
Is it the same error you faced last time?
As far as I remember the issue you faced was related to not having computing resources. Did you manage to run these models on a bigger GPU/CPU?
If you’re facing a different issue, please don’t hesitate to post the logs directly on the MONAI Label repository: Issues · Project-MONAI/MONAILabel · GitHub