There is a question about metal

[NERV-MAGI]

First of all, thanks for your contribution to the network research.

It works very well in most cases but we found some problem.
First,If a patient who Suffer from cerebral hematomas need to Wear Surgical head stand. there have some CT scans image before this patient's operation which shown two or three kind of metals that belonging to different materials. It is hard to apart them for creating mask.if we use the same way which like CNN-MAR to prepare train or test data. It will have some serious problems because that is totally differently about real image.

Could you give some proposal about what should we do for this?

Secondly, we found that when using natural images to make metal artifact images, some metals are not at the anatomical position they should be at and this situation could be found in lots of work and research.Even in CNN-MAR,they said that the mask should at the anatomical position they suit be at.

For training a model with a fixed relative position of metals, whether using masks will affect the network performance？

looking forward to your reply and wishing you success in your research.