CAT report and quality control (in development):

CAT now also prints image quality measurements (Dahnke et al. 2016) and major global subjects specific measures in addition to the used preprocessing parameters. These data are further saved in the "cat_*.xml" files for each subject that can be read by cat_io_xml.

Image quality measures

Due to the wide range of rating and marking systems, the rating was defined for a percentage range and a numerical scaling that allows mapping to nominal numbers or letters with chromatic variants ("+" and "-") shown in table below. The obtained quality ratings range from 0.5 (100 rating points (rps)) to 10.5 (0 rps) with values around 1 and 2 describing (very) good image quality (grad A and B) and values around 5 (grad E) and higher (grad F, < 50 rps) indicating problematic images. The scaling of the QMs was performed on the basis of theoretical aspects and the analysis of synthetic and real MRI data, especially the BWP (Brain Web Phantom; Collins et al. 1998).

Quality rating

The image quality measures describe the properties of the image before CAT12 processing. They were estimated by using the tissue segmentation and were (nearly) independent of subject sex and health status.

The resulting ratings were combined as weighted average image quality rating IQR (Dahnke et al. 2016). Tests of different databases such as ADNI, IXI, INDI, OASIS, PPMI etc. demonstrated that published data got good quality ratings in general. Please note that excellent rating is defined for extraordinary good images that can be achieved by averaging several ultra high resolution scans on high field systems (e.g. Winterburn et al. 2013), and that typical scientific (clinical) data is expected to get just good to satisfactory ratings.

Quality distribution of large projects

Low image quality can lead to underestimation of the GM in most preprocessings (Reuter et al. 2015, Dahnke et al. 2016). The following figure shows the effect of motion artifacts on IQR and the relative GM volume rGMV of a children dataset processed by 5 different tissue segmentations.

Quality rating

References:

< Methods