rickardcronholm / CTC_auto

A python module that converts DICOM CT data to voxalized phantom for the EGSnrc Monte Carlo code DOSXYZnrc
Other
8 stars 6 forks source link

Density resampling issue #6

Open inamoto85 opened 6 years ago

inamoto85 commented 6 years ago

Hi, thank you for sharing the code! Have you considered some improvements on the map_coordinates function? Since It's using a slice by slice interpolating scheme, there is no possibility to do resampling in the Z direction. I think ndimage.zoom could be a good candidate, or using directy RegularGridInterpolator.

rickardcronholm commented 6 years ago

Thanks for the feedback. I did consider that. There is a bug in ndimage.zoom when downsampling ( https://github.com/scipy/scipy/issues/7324) so I didn't want to use that. RegularGrid is only linear or nearest so I scrapped that too. You are correct that there is a limitation on z-direction but in my experience you typically want the same resolution in slice direction as the slice thickness.

On Friday, June 29, 2018, Hideki_Nakamoto notifications@github.com wrote:

Hi, thank you for sharing the code! Have you considered some improvements on the map_coordinates function? Since It's using a slice by slice interpolating scheme, there is no possibility to do resampling in the Z direction. I think ndimage.zoom could be a good candidate, or using directy RegularGridInterpolator.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/rickardcronholm/CTC_auto/issues/6, or mute the thread https://github.com/notifications/unsubscribe-auth/APEKdNEnve0GWQQtPlqrsc5d7HWaRWATks5uBdtpgaJpZM4U8lkq .

inamoto85 commented 6 years ago

Hi Rickard, Thank you for your response. I saw that the scipy.ndimage.map_coordinates can actually resample on higher dimension data (like something you did in the current code but including the third dimension coordinates-index interpolation). I have implemented something similar to this link, and the initial tests looks OK. Would you be interested to take a look?