Open AccessAmyloid burden quantification depends on PET and MR image processing methodology
Author(s) -
Guilherme Domingues Kolinger,
David Vállez García,
Antoon T. M. Willemsen,
Fransje E Reesink,
Bauke M. de Jong,
Rudi Dierckx,
Peter Paul De Deyn,
Ronald Boellaard
Publication year - 2021
Publication title -
plos one
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.99
H-Index - 332
ISSN - 1932-6203
DOI - 10.1371/journal.pone.0248122
Subject(s) - artificial intelligence , pattern recognition (psychology) , image processing , computer science , segmentation , parametric statistics , image segmentation , image registration , a priori and a posteriori , pipeline (software) , computer vision , mathematics , image (mathematics) , statistics , philosophy , epistemology , programming language
Quantification of amyloid load with positron emission tomography can be useful to assess Alzheimer’s Disease in-vivo . However, quantification can be affected by the image processing methodology applied. This study’s goal was to address how amyloid quantification is influenced by different semi-automatic image processing pipelines. Images were analysed in their Native Space and Standard Space ; non-rigid spatial transformation methods based on maximum a posteriori approaches and tissue probability maps (TPM) for regularisation were explored. Furthermore, grey matter tissue segmentations were defined before and after spatial normalisation, and also using a population-based template. Five quantification metrics were analysed: two intensity-based, two volumetric-based, and one multi-parametric feature. Intensity-related metrics were not substantially affected by spatial normalisation and did not significantly depend on the grey matter segmentation method, with an impact similar to that expected from test-retest studies (≤10%). Yet, volumetric and multi-parametric features were sensitive to the image processing methodology, with an overall variability up to 45%. Therefore, the analysis should be carried out in Native Space avoiding non-rigid spatial transformations. For analyses in Standard Space , spatial normalisation regularised by TPM is preferred. Volumetric-based measurements should be done in Native Space , while intensity-based metrics are more robust against differences in image processing pipelines.