Evaluating the sensitivity of Centiloid quantification to pipeline design and image harmonization
Mahnaz Shekari, David Vállez García, Lyduine E. Collij, Fiona Heeman, Núria Roé‐Vellvé, Santiago Bullich, Christopher Buckley, Frederik Barkhof, Gill Farrar, Hugh Pemberton, Juan Domingo Gispert,- Psychiatry and Mental health
- Cellular and Molecular Neuroscience
- Geriatrics and Gerontology
- Neurology (clinical)
- Developmental Neuroscience
- Health Policy
- Epidemiology
Abstract
Background
The Centiloid scale is a well‐established standardized metric for estimating amyloid load in clinical and research settings. In this study, we evaluated the stability of Centiloid to pipeline design options, tracer, and image harmonization.
Method
A total of 670 participants of the AMYPAD DPMS and PNHS multicenter studies with either [18F]flutemetamol or [18F]florbetaben PET scans and T1‐weighted MRI were included (Table1). PET images were harmonized to achieve an 8 mm Full‐Width‐at‐Half‐Maximum(FWHM) effective image resolution. Using SPM12, 32 Centiloid pipelines were created, calibrated and validated, based on combinations of four reference regions (Whole Cerebellum[WCB], Cerebellar Gray[CG], Whole Cerebellum+Brainstem[WCB+BSTM], Pons), two target VOI types (standard GAAIN vs subject‐based), two reference region types (standard GAAIN vs subject‐based) and two analysis spaces (native vs MNI). Generalized Estimating Equations (GEE) were used to evaluate the impact of the different factors on Centiloid. First, a base model including only the pipeline design factors was defined. Then, harmonization status (original vs harmonized PET images), tracer, and grey matter volume, as a measure of cerebral atrophy, were added to the base GEE model. Analyses were performed in the two cohorts separately and changes >5 Centiloid were considered relevant.
Result
Quantification space, target VOI, and reference region type had a small impact on Centiloid. Selection of reference region had the largest impact on Centiloid as using the Pons resulted in ∆Centiloid∼‐12.00 compared with using the standard reference region, WCB (Table2). Harmonization status affected Centiloid only for some reference regions: CG and Pons were sensitive to harmonization status (∆Centiloid∼ 5.45 and ‐5.70 respectively). In contrast, Centiloid was minimally affected by harmonization when using WCB and WCB+BSTM (Table3). Tracer and atrophy had a minimal effect on Centiloid for both DPMS and PNHS cohorts.
Conclusion
Centiloid quantification is robust against differences in analysis space, reference region type, target VOI, tracer and the presence of atrophy. However, choice of reference region can significantly influence Centiloid, with the Pons having the largest impact on Centiloid relative to other reference regions. Centiloid is stable against harmonization status and image resolution heterogeneities while using WCB as reference region.