LNDB pre-processing

[DrMachado]

Hello, first I would like to thank your team because I find your work very helpful and impressive.

I'm trying to reproduce the dataset and can't find the part relative to how the LNDB data was pre-processed, especially because this dataset is complicated and contains annotations from different radiologists. Could you be more explicit about the steps you took to prepare this dataset ? I am aware about the repackaging issue with the license of LNDB unfortunately.

Kinds regards,
Léo Machado

[MJJdG]

Hi Dr Machado,

I've looked up for you how we preprocessed the LNDb data. We summed the masks from the different radiologists for the nodule segmentations and included all voxels where > 1 radiologists had made an annotation. Then we ran our preprocessing pipeline to extract VOI's around each nodule. Since every dataset has different annotation guidelines and readers with different experience levels it's often hard to determine an optimal way to combine masks like this. Based on our checks this approach yielded good results for this dataset. Let me know if you have further questions,

Kind regards,

Max