Use of pre-trained models

Question: "I wonder if I am allowed to use ImageNet pretrained model as a backbone for the segmentation model?"

Answer: We do not allow participants to use pre-trained models because participants could use existing models pre-trained on T2 images to perform cochlea and VS segmentation. Of course, it wouldn't be the case with the ImageNet model, but we want to have a single rule regardless of the type of pre-trained model.

Question: "I just wanted to check if external algorithms could be used for pre-processing tasks even though they are data-driven and trained on a separate dataset (e.g. a deep learning framework for skull stripping used as is)."

Answer: We do not allow the use of pre-trained models since there are publicly available models for VS segmentation on T2 scans. Were you planning to use HD-BET? You could use ROBEX instead.