NSCLC-Radiomics collections - what annotations are most crucial?

I am looking at this dataset NSCLC-Radiomics and I see that it has 4 classes - tumor, lung, heart, esophagus.
Are there any additional classes that could be relevant? Like - trachea, bronchus, mediastinum (which houses the heart)… could these be useful for metastasized cases?
Is there value in segmenting them?

The classes of tissue that are relevant will be defined by the use case / application. A structure that is critical for one application may not be that critical for another application. I am afraid there is no general answer to your question.

I agree with @fedorov.

However, not sure if you’ve checked, but in IDC v13 there are some additional annotations for NSCLC-Radiomics that may be useful for you. These include the heart, esophagus, trachea and aorta for 414 out of 422 patients. These were segmented using three nnU-Net architectures (2d, 3d low res, 3d high res) with test time augmentation (using pre-trained models from Task055 SegTHOR).

In the portal, under analysis results choose ‘nnU-Net-BPR-annotations’.