Hello, we have set up a
clara_train_covid19_annotation_ct_lung on a DGX V100 machine and tested it through the 3D Slicer software.
We tried to annotate a volumetric CT image using dextr3d on the left side of a lung, but the output resulted a masking on the right side. And so did when it was annotated on the reversed side.
Is this a common problem where the model is currently being developed, or did we make a technical mistake?
Thanks for your interest in clara train AIAA.
clara_train_covid19_annotation_ct_lung model is designed to segment both lungs left and right not the diseased areas. This is mainly to have a ROI for the classification model on NGC https://ngc.nvidia.com/catalog/models/nvidia:med:clara_train_covid19_3d_ct_classification
In other words if the segmentation model is giving bad results you should use this annotation model to annotate the full lungs then retrain the lung segmentation model.
Hope that helps