Head & Neck segmentation

[MattAWard]

Hi are there any plans to facilitate H&N segmentation using TotalSegmentator? If not, is it possible to train our own model using marked-up CT data and their corresponding RTSTRUCT files?

[wasserth]

What exactly do you mean by head and neck segmentation? What structures would you like to have segmented?

[MattAWard]

This would be for segmentation relating to head and neck radiotherapy. Regions of interest typically include:

• Brain
• Brainstem
• Eyes
• Eye Lens
• Optic Chiasm
• Optic Nerves
• Cochlea
• Parotids
• Oral Cavity
• Mandible
• Spinal Canal/ Cord
• Oesophagus
• Thyroid Gland
• Brachial Plexus
• Carina

Thanks.

[MattAWard]

Any idea about this? As I said, an ability to train a model using our own anonymised CT data sets with these structures outlined would also be a great solution.

[wasserth]

Training a nnUNet should not be too difficult. You have to convert your data to nifti and follow the instructions of the nnU-Net Documentation (https://github.com/MIC-DKFZ/nnUNet). Having such a model would be great and then we could use this to help in annotation of the TotalSEgmentator dataset and then the TotalSegmentator could also do this.

[MattAWard]

That's great thanks @wasserth I'll look into training a model with some of our data and then we can take it from there.