Clarification of the input data X and Y for Centreline prediction

[Leandroscholz]

Hi @giesekow congratulations for the work! DeepVesselNet appears to be a great tool and I am very excited to test it! I am just starting to study and understand Deep Learning tools and I would like to ask more clarification about the inputs of DVN so I could use if for centreline prediction.

I understood that X are the raw images and Y are annotated images (Ground Truth) but it is not clear what the dimensions depict. In your example.py :

X.shape
(10,1,64,64)
Y.shape]
(10,2,64,64) 

If I wanted to train the network with 3D images (say n=100 images) of 124x124x124 px and Y centreline annotations (Assuming then X.shape (100,1,124,124,124)), what steps do I need to take? Your paper in arXiv tells the DVN has to be trained on the binary mask first because it uses the networks use the probabilistic segmentation masks and it is not clear what to do after training the DVN-FCN for vessel segmentation.

Also, do you happen to have a snapshot of the trained model for DVN-FCN with the Synthetic Dataset provided?

Cheers,
Leandro S.