Find good strategy to 'hide' the preprocessing at inference time

[jcohenadad]

Currently, data need to be preprocessed in order to run the model. This preprocessing includes cord segmentation, bounding box creation, etc.

This adds complications for the end-user. Ideally, the preprocessing would be incorporated in the inference script, the same way it is done for the tumor segmentation or sct_segment_lesion.

One possibility would be to train another model for finding the bounding box, and then create a cascaded workflow of CNNs.