We want to demonstrate, as part of this study, that the more SC slices we include, the more sensitive we are to catch atrophy. Changes on "csa_rescale_stat.py" try to integrate this through: computing and plotting mean error of CSA between GT and measured for different vertebrae levels (C2:C3, C2:C4, C2:C5).
"sct_resample" did not rescale the image to simulate atrophy. The image voxel number was reduced only to increase the voxels size leaving an image of same size but poorer quality. The new script "affine-rescale.py" directly changes voxel size by multiplying the affine array by the desired rescaling coefficient. My understanding is that every image of openneuro is 256x256x51.2 mm^3 and that voxel/(subject or scanner?) ratio is given on the affine array diagonal.
Results are still in progress and do not correspond to what is expected. Error on even small rescaling remains important. Maybe a deeper look into deepseg functioning for rescaling and resampling changes could be a good next step.
We want to demonstrate, as part of this study, that the more SC slices we include, the more sensitive we are to catch atrophy. Changes on "csa_rescale_stat.py" try to integrate this through: computing and plotting mean error of CSA between GT and measured for different vertebrae levels (C2:C3, C2:C4, C2:C5).
"sct_resample" did not rescale the image to simulate atrophy. The image voxel number was reduced only to increase the voxels size leaving an image of same size but poorer quality. The new script "affine-rescale.py" directly changes voxel size by multiplying the affine array by the desired rescaling coefficient. My understanding is that every image of openneuro is 256x256x51.2 mm^3 and that voxel/(subject or scanner?) ratio is given on the affine array diagonal.
Results are still in progress and do not correspond to what is expected. Error on even small rescaling remains important. Maybe a deeper look into deepseg functioning for rescaling and resampling changes could be a good next step.
DONE: