Intensity normalization

[wftubby]

According to your latest paper: I have some question？

1. We compute following properties for each patient/case and use properties per patient to normalize data. We use ground truth mask to compute foreground pixels for all classes to compute mean, std, percentile_99_5 and percentile_00_5.Would you tell me these 'data(mean, std, percentile_99_5 and percentile_00_5)' was computed before 'crop' or after 'crop'?
2. if the type of data is PET image? How to normalize data?

[FabianIsensee]

After crop, but this should make no difference.

PET image will be handled like any other non-CT image: each sample is normalized with its own mean and std

[wftubby]

thanks for your reply,
and the mean and std were computed based on the intensities of the whole images or belonging to any of the foreground classes?
about patch size , one patch size in one patient case or multi-patch size in one patient case?

[FabianIsensee]

The details are in the paper ;-)

only CT uses foreground classes. Other modalities use the whole image

[wftubby]

thank you

[wftubby]

PET image will be handled like any other non-CT image: each sample is normalized with its own mean and std which computed from their foreground or the whole images?

[FabianIsensee]

how would it use the mask on test cases? That is impossible... there is no mask for those

So it's just using the entire image