PhD defence: Martin Nørgaard, MSc, 'Optimizing Preprocessing Pipelines in PET/MR Neuroimaging'
Friday, 3. May 2019, 14:00 - 17:00
Hits : 444
by dorthe
Auditoriet opgang 61A, Rigshospitalet, Henrik Harpestrengs Vej, 2100 København Ø

After the defence, the Neurocenter will host a reception at the Neurobiology Research Unit, Rigshospitalet, Juliane Maries Vej 28, 3rd floor


Assessment committee: Professor Liselotte Højgaard, University of Copenhagen (Chair), Professor Ronald Boellaard, VU University Medical Center, Amsterdam, NL, Professor R. Todd Ogden, Columbia University, New York, USA

Academic advisors: Principal supervisor: Professor Gitte Moos Knudsen, University of Copenhagen, Primary co-supervisor: Senior scientist Claus Svarer, PhD, Rigshospitalet, Co-supervisors: Asst. Professor Melanie Ganz, University of Copenhagen and Professor Stephen C. Strother, University of Toronto.


Positron Emission Tomography (PET) is a state-of-the-art imaging technique for measuring the spatial distribution of neurotransmitters and receptors in the living human brain. However, the PET signal is influenced by complex spatio-temporal noise patterns arising from sources of radioactive decay, head motion and scanner- specific limitations.
A large set of preprocessing algorithms have been developed to remove various sources of noise, but there is currently a limited consensus in the literature on the most optimal preprocessing strategy. Furthermore, it is not well understood how the choice of preprocessing strategy may affect the variability of the data and ultimately the conclusions of a study. This thesis develops a framework for the evaluation of preprocessing performance in PET using the radioligand [11C]DASB, targeting the serotonin transporter, as exemplary case. In the five included research papers, I evaluate current preprocessing strategies in the literature, how they affect measures of test-retest bias, variability and false-positive rates, and how they may lead to different conclusions in a double blind, randomized, placebo-controlled study. Finally, I provide a statistical framework for adequately controlling the false-positive rate when dealing with large sets of preprocessing options.
In this work, I show that (1) variations in choice of preprocessing strategy are an overlooked aspect in modern PET neuroscience, (2) measures of bias, within- and between-subject variability are significantly affected by preprocessing strategy, and significant differences between test and retest were obtainable despite correcting for multiple comparisons and (3) different preprocessing strategies lead to different neurobiological conclusions. My findings suggest that the preprocessing stage contributes with considerable variance into the data, with the preprocessing steps motion correction, partial volume correction and kinetic modeling contributing the most. I show that knowledge about the variability of preprocessing is critical to limiting false-positive rates. This underlines the importance of selecting preprocessing strategy with great caution. Finally, I present my view on future directions and best practices for handling preprocessing variability across PET centres.