Are the intensities values in MRI Nifti images always have the same distribution (or statistics)? For instance, the highest intensity is always around 6000. Or does it depend on the instrument used to capture the MRI image? My understanding is MRI Nifti images always have intensities values encoded in 32-bit unsigned integer. Is that correct? Any references that discuss the variety of the intensities for different MRI images captured from the same instrument or other instrument is appreciated. Ideally, I hope to see that T1W Nifti images always have a unified mean and standard deviation.
None of that is true. The units are arbitrary, and the T1-weighted data comes in a lot of different ranges though often in the hundreds or thousands. The data are often stored in 16-bit signed or unsigned integer, but you can find 8-bit, 10 or 12-bit in 16-bits, 32-bit floating point. Computed templates, averages and other processed data can have almost any range, e.g. 0.0-1.0. Within AFNI, you can use a program like 3dUnifize to standardize a T1w dataset to some degree. That tries both to remove non-uniformity within a dataset, but it also scales the data to be similar.
1 Like
Perfect answer! Thank you for 3dUnifize suggestion.
@dglen provides an excellent answer. For completeness, raw voxel intensity in most MR modalities is relative and not calibrated. However, there are exceptions like ASL. In contrast, raw CT and Xray tend to report calibrated Hounsfield units. Many derived images for CT, MR, PT are calibrated values (e.g. MTT, CBF, CBV, etc). Older MR equipment used 12-bit ADC, so raw voxel values were in the range 0…4095, while 16-bit ADC is common now, with unsigned magnitude images stored in the range 0…65535.
2 Likes