To prepare an initial dataset I need to do annotations (segmentation masking or bounding box) around the spinal cord area. Can I use the SCT for this purpose? What do you recommend? ITK SNAP or slicer can be also considered appropriate choices for this purpose? (considering that the sc area is a little bit tricky and has several levels)
Secondly, to evaluate the quality of acquired images, what do you suggest doing? Although the radiologist follows the generic protocol, the image quality does not seem great in some modalities for example in DWI, I tried to analyze them using SCT and check the qMRI measurements. Is this approach correct? What do you usually do for making sure that the quality of images is good enough?
To prepare an initial dataset I need to do annotations (segmentation masking or bounding box) around the spinal cord area. Can I use the SCT for this purpose? What do you recommend? ITK SNAP or slicer can be also considered appropriate choices for this purpose? (considering that the sc area is a little bit tricky and has several levels)
it depends what you mean exactly by “annotation (segmentation masking or bounding box)”. If you meant “manually correcting for incorrect automatic segmentation”, then you can indeed use any NIFTI editor, like ITKsnap or FSLeyes. Please see the procedure described in spine-generic to go through the manual corrections and re-use them for a second pass of the analysis. You should create a script that loops across subjects instead of manually clicking on each image correct.
Regarding the bounding box, this is something you can easily do with sct_label_utils. Eg:
sct_crop_image -i t2.nii.gz -g 1 -b 0 -o mask.nii.gz
# note: we use "-g 1" to get the viewer and "-b 0" to create a mask (instead of cropping the image)
Secondly, to evaluate the quality of acquired images, what do you suggest doing? Although the radiologist follows the generic protocol, the image quality does not seem great in some modalities for example in DWI, I tried to analyze them using SCT and check the qMRI measurements. Is this approach correct? What do you usually do for making sure that the quality of images is good enough?
Checking qMRI measurements introduces a bias. Image quality should be assessed before computing qMRI metrics. E.g spotting for poor fat sat, motion, Gibbs, etc. I usually do the QC using SCT’s HTML QC module (see the spine-generic link above for video tutorial).
Thanks very much, @jcohenadad
I am using ITKsnap these days, I might try FSLeyes as well if the former was not satisfying.
About quality assessment, I usually used to obtain the qMRI measurements, but checking the quality of the image as you mentioned needs the specialty of radiologists… So, after converting the images to NIFTI we first assess the quality through visual signs in the image itself.
but checking the quality of the image as you mentioned needs the specialty of radiologists… So, after converting the images to NIFTI we first assess the quality through visual signs in the image itself.
If you are working on comparing the performance of MRI biomarkers, assessing the quality of the input data (ie: DICOM/NIFTI before processing) is a skill you absolutely need to acquire. You cannot rely only on radiologist to tell you if an image is PASS/FAIL for your analysis pipeline. It takes time, but it should be part of your training IMHO.