Converting dicom images to nifti file

HI, got the question~
I want to use (clara_ct_seg_liver_and_tumor_amp) segmentation model to predict , but Scores is not better than web show scores.
I did not use the [nvmidl-dataconvert] you provided, I used github’s method [GitHub - icometrix/dicom2nifti] to convert.
Is it affected because of this?
Thx~

Hi
Thanks for your interest in Clara Train SDK.
I think dicom2nii tool works for conversion however it doesn’t resample the data to 1x1x1 resolution required by the model. For this set you would need to resample either with our tool or something else
Also could you expand on “but Scores is not better than web show scores” what is Not better and what is the web scores ?

Hope this helps

I mean I input my nii file to model and get predict result is not good.
Your website{NVIDIA NGC} show scores is Liver: 0.924, but I check I result that I can’t see complete liver.

Hi
If your input nii has different resolution than 1x1x1 then the result would be way off. Please resample your nii with our tool. You can also use the AIAA with slicer, which would do that resampling for you

If you still get bad results then may be your nii is flipped / orienting differently than our training data.

Can you add some screenshots, may be we can spot why this happens
Thanks