MAISI Inference

I am trying to use MAISI inference and I have a couple of questions.

my pipeline now to generate a segmentation mask to give as input for MAISI inference is:

1. generate segmentation mask using TotalSegmentatorV2
2. generate segmentation mask using Visat3D
3. unify labels (convert TS labels to same values of Vista3D)
4. generate unified mask to take best segmentation
5. use MAISI inference

my questions are:

1. is that pipeline helps to get best results or it has some problems, as I could not neither find in the paper nor documentations how you prepare the segmentation masks.
2. Important question: both segmentation networks do not provide segmentation masks for air or body regions. I investigated the training dataset and found that the value 200 is assigned to those parts, however I could not find anywhere what is HU threshold you are using, so please also if you can guide onto that part whether where to find it or what is the threshold being used. so if you can advise how you prepare the body segmentation mask with value 200 and integrate with the segmentation masks from the segmentation networks.

Inizia con il workflow di inferenza MAISI descritto nell’issue, quindi confronta gli output di TotalSegmentatorV2 e Visat3D con il dataset di training. Documenta come vengono unificate le label, come il valore 200 viene assegnato alle regioni di aria o del corpo e quale soglia HU viene utilizzata; il lavoro è completato quando la procedura di preparazione delle mask è sufficientemente chiara da poter essere riprodotta.