Abstract

Although normative modelling has been recently widely adopted by the neuroimaging community to estimate deviation of cohorts or single subjects, from a reference population's trajectory, it has never been applied on molecular neuroimaging datasets. This is because the typical sample size of molecular imaging datasets of single studies is not adequate for a reliable estimation of population models. Here, we pooled scans from different datasets and compared the performance of common neuroimaging harmonisation techniques when used on molecular neuroimaging data, by measuring the effect of the harmonisation on the deviation scores of the normative model. As harmonisation methods we employed a 3D Gaussian filter and a Bayesian scale and shift estimation method (Combat). By statistically testing the parameters of the deviation-score distribution and inspecting the explained variance map of the model, we selected Combat, trading-off between the two parameters' performance, as the most suitable way to harmonise multi-site/multi-scanner molecular neuroimaging datasets.

Original languageEnglish
Title of host publication8th National Congress of Bioengineering, GNB 2023 - Proceedings
PublisherPatron Editore S.r.l.
ISBN (Electronic)9788855580113
Publication statusPublished - 2023
Event8th National Congress of Bioengineering, GNB 2023 - Padova, Italy
Duration: 21 Jun 202323 Jun 2023

Publication series

NameConvegno Nazionale di Bioingegneria
ISSN (Electronic)2724-2129

Conference

Conference8th National Congress of Bioengineering, GNB 2023
Country/TerritoryItaly
CityPadova
Period21/06/202323/06/2023

Keywords

  • Dopamine
  • Multisite Data Pooling
  • Normative Modelling
  • PET Imaging

Fingerprint

Dive into the research topics of 'Assessing Multisite PET Neuroimaging Harmonisation for Normative Modelling'. Together they form a unique fingerprint.

Cite this