Premium
Harmonization of amyloid PET scans minimizes the impact of reconstruction parameters on centiloid values
Author(s) -
Shekari Mahnaz,
NiñerolaBaizán Aida,
Salvadó Gemma,
Battle Mark R,
Buckley Christopher,
Farrar Gill,
Operto Greg,
Falcon Carles,
ArenazaUrquijo Eider M,
Perissinotti Andrés,
Molinuevo Jose Luis,
Gispert Juan Domingo
Publication year - 2020
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1002/alz.045294
Subject(s) - imaging phantom , nuclear medicine , positron emission tomography , neuroimaging , artificial intelligence , computer science , mathematics , medicine , psychology , neuroscience
Background The Centiloid method has been proposed to enhance the comparability of amyloid PET quantification across different amyloid tracers and to provide a reference metric. In this study, we first quantified the impact of reconstruction parameters on Centiloid values for different reference regions, then we evaluated the reduction of this source of bias after PET image harmonization. Method A total of 71 cognitively unimpaired participants from ALFA+ cohort were included in this study. For each participant a 18 F‐flutemetamol PET scan and a T1‐weighted MRI scan were acquired. A Hoffman 3D‐brain phantom filled with 0.5 mCi of 18 F‐FDG was scanned in the same tomograph. Participant and phantom PET data were reconstructed with 4 different protocols used in the clinical and research settings (Table1). The reconstruction with the coarsest spatial resolution was selected as the reference. The remaining 3 were harmonized to yield the same effective spatial resolution using the protocol used in the ADNI study. All images were quantified with a validated Centiloid pipeline using an AAL composite and different reference regions (Whole Cerebellum, Cerebellar Gray, Pons and White Matter). The mean absolute difference in Centiloids (CL) between different reconstructions of the same scan was calculated using the coarsest resolution as reference. Differences below 2.5 CL were considered to be irrelevant based on published test‐retest data. Result Respectively, Figures 1 and 2 show the Hoffman phantom and the amyloid PET scans before and after harmonization. While mean absolute differences were below the test‐retest threshold for the Cerebellar Gray and whole cerebellum, differences up to 9 CL were observed for the Pons (Table2). White matter as reference region showed the biggest sensitivity to reconstruction parameters and the bias was dependent on the amyloid load (Figure 3c). After harmonization differences were always below the test‐retest threshold. Conclusion Reconstruction parameters impact the Centiloid values as a function of reference region. While the impact using Cerebellar gray and the Whole Cerebellum was minor, important differences were observed when using the pons and white matter as reference regions. In all cases, harmonization of the effective spatial resolution was able to bring these biases below the test‐retest threshold.