Non-standard pipeline without MRI has replicability in computation of Centiloid scale values for PiB and 18F-labeled amyloid PET tracers

Abstract

The magnetic resonance imaging (MRI)-less non-standard pipeline for amyloid positron emission tomography (PET) published by Bourgeat et al., in 2018 calculates Centiloid scale values that are highly consistent with those computed using the standard pipeline. The purpose of this study was to demonstrate that the non-standard pipeline can compute Centiloid scale values in high agreement with the standard pipeline when using different datasets of amyloid PET tracers in our local computer environment. PET images of ¹¹C-Pittsburgh compound B (¹¹C-PiB), ¹⁸F-florbetapir, ¹⁸F-flutemetamol, ¹⁸F-florbetaben, and ¹⁸F-NAV4694 from the calibration dataset were processed using both the standard and non-standard pipelines, and the computed cortical standardized uptake value ratio (SUVr) value was converted to the Centiloid scale value using the method described by Klunk et al., in 2015. The conversion equations from the SUVr to Centiloid scale values for each tracer were obtained during this process. Using these equations, we compared the Centiloid scale values obtained using the standard and non-standard pipelines using the validation datasets of each tracer from the Japanese Alzheimer's Disease Neuroimaging Initiative (J-ADNI), Alzheimer's Disease Neuroimaging Initiative (ADNI), and Australian Imaging Biomarkers and Lifestyle (AIBL). In the calibration datasets, there was high agreement (R² > 0.97) and slight bias between the Centiloid scale values calculated by the non-standard and standard pipelines for all tracers. Despite relatively little NAV4694 data in the validation datasets, there was high agreement between the Centiloid scale values calculated using the non-standard and standard pipelines for all tracers. The bias for florbetaben and NAV4694 using the non-standard pipeline was 1.6% underestimation and 3.3% overestimation, respectively; these values were smaller than those reported by Bourgeat et al. Analysis of outliers also suggested that the non-standard pipeline might be vulnerable to anatomical anomalies. Given the slight variance of the Centiloid scale in young controls, flutemetamol and NAV4694 might be suitable tracers for the non-standard pipelines. This study demonstrates the replicability of the non-standard pipelines across computing environments, datasets, scanners, and tracers. When MRI is not available, the non-standard pipeline may provide information to aid in visual assessment of amyloid PET.

Keywords: Centiloid; Florbetaben; Florbetapir; Flutemetamol; NAV4694; Pittsburgh compound B.

Fig. 1

Flow chart of the non-standard pipeline. Each PET image is affinely registered to the mixed template using reg_aladin. The NMI is then computed between the affinely registered image and the adaptive template. The adaptive template is generated by linearly combining $A_{neg}$ and $A_{pos}$. The NMI is maximized using Powell’s optimization method. Finally, the affinely registered image is nonlinearly registered to the adaptive template using reg_f3d. The SUVr is computed using the cortical and whole cerebellar VOIs for the nonlinearly registered image. Abbreviations: $A_{neg}$, PiB-negative template; $A_{pos}$, PiB-positive template; NMI, normalized mutual information; VOI, volume of interest.

Fig. 2

Correlation of the Centiloid scale values from the GAAIN PiB dataset between the SPM8 standard and non-standard pipelines.

Fig. 3

Correlation of the Centiloid scale values from each GAAIN ¹⁸F-labeled tracer dataset between the SPM8 standard and non-standard pipelines. The x-axis shows the Centiloid values from the SPM8 standard pipeline, and the y-axis shows the Centiloid values from the non-standard pipeline for FBP (a), FMM (b), FBB (c), and NAV (d).

Fig. 4

Correlation of the Centiloid scale values from the J-ADNI PiB (a), ADNI FBP (b), AIBL FMM (c), ADNI FBB (d), and AIBL NAV (e) datasets between the SPM8 standard and non-standard pipelines. The x-axis shows the Centiloid values from the SPM8 standard pipeline, and the y-axis shows the Centiloid values from the non-standard pipeline.

Fig. 5

An example of failure in anatomical normalization using the non-standard pipeline. (a) Coronal section of the template image. (b) Anatomically normalized target PET image in the template space. Note that the center of the crosshairs is positioned on the same coordinates as (a). (c) Axial section of the template image with the overlaid translucent white cortical VOI. (d) Axial section of the anatomically normalized target PET image in the template space with the overlaid translucent white cortical VOI. A registration error between the template (a) and nonlinearly registered image (b) is found in the high convexity (the center of the crosshairs). This registration error could be due to the high uptake of a tracer in the diploe and the subcutaneous tissue of the scalp in the target image, leading to failure of cortical VOI extraction (d).

Fig. 6

Another example of failure in anatomical normalization using the non-standard pipeline. (a) Axial section of the template. (b) Anatomically normalized target PET image in the template space. Note that the center of the crosshairs is positioned on the same coordinates as (a). (c) Axial section of the template image with the overlaid translucent white whole cerebellar VOI. (d) Axial section of the anatomically normalized target PET image in the template space with the overlaid translucent white whole cerebellar VOI. A registration error between the template (a) and nonlinearly registered image (b) is found in the posterior edge of the bilateral cerebellar hemispheres (the center of the crosshairs). This was due to the expanded cerebrospinal fluid space, leading to failure of whole cerebellar VOI extraction (d).