ASL data conversion


This paragraph discusses the specific scaling issues for GE, Philips, and Siemens. Beware that these issues are occurring at the time of the release of version 1.5.0, and might be resolved in the future.


GE applies an RF amplifier factor that should be multiplied by the number of signal averages (NSA) as GE sums instead of averages them.


Typically, Philips ASL datasets use the standard DICOM rescale tags only to rescale data for viewing, whereas private rescale tags are used for scaling to acquisition values needed for proper quantification.

The procedure for using both the scaling types is described previously [reference required], although the authors did not mention that three DICOM fields are typically used for this purpose - (2005, 100e), (2005, 110e), or (2005, 120e).

It may also occur that the RescaleSlope (0028, 1053) is set to 1 and RescaleSlopeOriginal (2005, 140a) needs to be used instead. Note that DICOM to NIfTI conversion software may omit the private slopes, in which cases the private slope needs to be applied after running the conversion.

An exception is the dcm2nii(X) software that does take private scale slopes into account in most cases. The dcm2nii(X) versions released between 2018 and 2019-09-02 can correctly scale the NIfTI according to the private scale slopes - if run with parameter -p - and indicates this by setting UsePhilipsFloatNotDisplayScaling==1 in the accompanying json file.

Two exceptions apply, dcm2nii(X) does not support DICOM tags (2005, 110e) and (2005, 120e) that still need to be read and applied additionally. Furthermore, the field RealWorldValueSlope (0040,9225) is sometimes provided as an alternative to RescaleSlope (0028,1053). In this case, dcm2nii(X) outputs a field PhilipsRWVSlope and sets UsePhilipsFloatNotDisplayScaling==1. This is deceiving because the standard RescaleSlope is applied instead of private-field scale slopes and this still needs to be corrected manually.

Note that this last-mentioned behavior is corrected in the newest dcm2nii version and general caution is recommended when dealing with Philips scale slopes. Also, Philips has recently implemented the option to apply these scaling factors upon export. The scaling of acquired/reconstructed ASL images may differ between control, label, and M0 images.


Several implementations of Siemens ASL WIP have a scaling between ASL and M0. This can be only modified and obtained at the scanner console and it is set to 10 by default.


The so-called PIXDIM error appears to be specific for files with a single repetition, more specifically the separately acquired m0scan. In the NIfTI header, dim[0] defines the number of dimensions, dim[i] defines the length of the i’th dimension, pixdim[i] specifies the voxel width along dimension i. For images with a single repetition, where pixdim[4] specifies the temporal width, the BIDS validator requires that the image is defined as 4D with dim[0] == 4, even though the image has a single repetition dim[4] == 1. Defining (dim[0] == 3) for a 44D volume with a single repetition (for instance: a 3D image) is common and correct, however, it is considered invalid by the validator.