Accepted Manuscript Implementing the centiloid transformation for using CapAIBL 11 C-PiB and β-amyloid 18 F-PET tracers Pierrick Bourgeat, Vincent Doré, Jurgen Fripp, David Ames, Colin L. Masters, Olivier Salvado, Victor L. Villemagne, Christopher C. Rowe PII: S1053-8119(18)30742-0 DOI: 10.1016/j.neuroimage.2018.08.044 Reference: YNIMG 15205 To appear in: NeuroImage Received Date: 7 March 2018 Revised Date: 9 July 2018 Accepted Date: 17 August 2018 Please cite this article as: Bourgeat, P., Doré, V., Fripp, J., Ames, D., Masters, C.L., Salvado, O., Villemagne, V.L., Rowe, C.C., the AIBL research group, Implementing the centiloid transformation 11 18 for C-PiB and β-amyloid F-PET tracers using CapAIBL, NeuroImage (2018), doi: 10.1016/ j.neuroimage.2018.08.044. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. ACCEPTED MANUSCRIPT Implementing the centiloid transformation for 11C-PiB and β-amyloid 18FPET tracers using CapAIBL Pierrick Bourgeat1, Vincent Doré1,2, Jurgen Fripp1, David Ames3, Colin L. Masters4, Olivier Salvado1, 1. CSIRO Health and Biosecurity, Brisbane, Australia RI PT Victor L. Villemagne2,4,5, Christopher C. Rowe2,5 and the AIBL research group SC 2. Department of Molecular Imaging, Austin Health, Melbourne, Australia 3. National Ageing Research Institute, Parkville, Victoria, Australia. Victoria, Australia. M AN U 4. The Florey Institute of Neuroscience and Mental Health, University of Melbourne, Parkville, 5. Department of Medicine, University of Melbourne, Melbourne, Australia TE D Corresponding author: Pierrick Bourgeat Level 5 UQ Health Sciences Building, Royal Brisbane and Women's Hospital, Herston, Queensland EP 4029 Australia. AC C Tel:(+61)732533659 Email:Pierrick.Bourgeat@csiro.au ACCEPTED MANUSCRIPT Abstract The centiloid scale was recently proposed to provide a standard framework for the quantification of β-amyloid PET images, so that amyloid burden can be expressed on a standard scale. While the framework prescribes SPM8 as the standard analysis method for PET quantification, non-standard methods can be calibrated to produce centiloid values. We have RI PT previously developed a PET-only quantification: CapAIBL. In this study, we show how CapAIBL can be calibrated to the centiloid scale. Methods Calibration images for 11C-PiB, 18F-NAV4694, 18F-Florbetaben, 18F-Flutemetamol and 18F- SC Florbetapir were analysed using the standard method and CapAIBL. Using these images, both M AN U methods were calibrated to the centiloid scale. Centiloid values computed using CapAIBL were compared to those computed using standard method. For each tracer, a separate validation was performed using an independent dataset from the AIBL study. Results Using the calibration images, there was a very strong agreement, and very little bias TE D between the centiloid values computed using CapAIBL and those computed using the standard method with R2 > 0.97 across all tracers. Using images from AIBL, the agreement was also high with R2 > 0.96 across all tracers. In this dataset, there was a small underestimation of the centiloid EP values computed using CapAIBL of less than 0.8% in PiB, and a small over-estimation of 1.3% in AC C Florbetapir, and 0.8% in Flutemetamol. There was a larger overestimation of 8% in NAV images, and 14% underestimation in Florbetaben images. However, some of these differences could be explained by the use of different scanners between the calibration scans and the ones used in AIBL. Conclusion The PET-only quantification method, CapAIBL, can produce reliable centiloid values. The bias observed in the AIBL dataset for 18F-NAV4694 and 18 F-Florbetaben may indicate that using different scanners or reconstruction methods might require scanner-specific adjustments. ACCEPTED MANUSCRIPT Keywords: Alzheimer's disease; Amyloid Imaging; Centiloid 1. Introduction β-amyloid (Aβ) imaging using PET has become the de facto standard to identify one of the RI PT neuropathological signs of Alzheimer’s disease (AD) in vivo. There is however great variability in the way Aβ burden is quantified from PET images, with each research centre developing their own pipeline to analyse the images, and using their own volume of interest (VOI) to compute the mean SC uptake in the neocortical and reference regions. This leads to great variability in the resulting numbers (1), and makes comparison across sites or studies difficult (2). This also means that M AN U pooling data across studies requires all images to be reanalysed to minimise bias. The variability is also exacerbated by the use of different tracers, each having different pharmacokinetics, binding potentials, recommended acquisition protocol and reference region (3,4,5,6). There are currently 5 different tracers that are commonly used in research studies, namely 11C-PiB (PiB), 18F- NAV4694 TE D (NAV), 18F-Florbetaben (FBB), 18F-Flutemetamol (FLUTE) and 18F-Florbetapir (FBP), with each tracer having their own recommended acquisition protocol, reference region and cut-off value for Aβ positivity. However, due to the variability from the quantification method, these cut-off values EP often need to be adjusted at each site so that they match visual reading. AC C The centiloid (CL) scale was developed by an international working group to alleviate some of these issues and provide a framework to standardise measures of Aβ burden from PET images (7). The framework allows any tracer or quantification method to be linearly mapped to the same scale: the centiloid scale. In this scale, 0 represents the typical Aβ burden in young controls, and 100 the typical Aβ burden in mild AD patients. A standard quantification pipeline based on SPM8 is prescribed to perform the initial anchoring of the centiloid scale, and calibration of new tracers or methods. To promote wide adoption of the centiloid scale, the VOI masks required for the ACCEPTED MANUSCRIPT quantification and the images needed for the calibration are all made freely available on the GAAIN website. As part of this effort, the calibration images and associated equations have already been released for 11C-PiB (7), 18F- NAV4694 (8), 18F-Florbetaben (9), 18F-Flutemetamol (10) and 18F- Florbetapir (11). RI PT One of the limitation of the standard SPM8 pipeline is that each PET image requires a corresponding MR image to provide anatomical constrain during the spatial normalisation to the common template. This can be an issue when MR imaging is impossible (claustrophobia, metal SC implants, non-compliant subject…), as this would prevent the automatic quantification of the M AN U accompanying PET image. For this reason, we have developed a PET-only approach (Computational Analysis of PET from AIBL or CapAIBL for short) that provide SUVR quantification of Aβ PET images without the need of a corresponding MR image (12,13). In this study, we apply the centiloid framework to the calibration of CapAIBL, and compare centiloid value derived from EP 2. Materials and Methods TE D all 5 tracers to those computed using the standard method. 2.1. Standard SPM pipeline AC C Each subject’s MRI was spatially normalised to the MNI-152 template using the SPM8 unified segmentation method. Each subject’s PET was then co-registered to their corresponding MRI, and spatially normalised to the template using the deformation parameters calculated from the MRI. For each subject, the mean retention within the standard VOIs from the GAAIN website, namely global cortical target (CTX) VOI and whole cerebellum (WC) VOI, was computed. The CTX/WC ratio was used to define the subject’s standardised uptake value ratio (SUVR). Since the whole cerebellum was found to be the reference region that gave the smallest standard deviation in both ACCEPTED MANUSCRIPT AD and young controls (YC), and lead to the largest effect size (7), this is the only reference region that will be considered in this study, and every mention of the SUVR will refer to the neocortical SUVR normalised by the whole cerebellum. RI PT 2.2. PET-only SUVR quantification using CapAIBL When no MR is available, PET images can be spatially normalising to a mean PET template directly, as is commonly done with FDG. However, when performed using Aβ PET images, the spatial SC normalisation introduces a bias in the resulting quantification, with an overestimation of SUVR in M AN U Aβ negative subjects, and an under-estimation in Aβ positive subjects (14). This effect can be reduced by using an adaptive template that matches its retention to that of the subject’s PET (12). The adaptive template is a linear combination of an Aβ negative (Aneg) and Aβ positive (Apos) template, with a weight w optimised by maximising the normalised mutual information between TE D the adaptive template and the target: A( w) = w * Aneg + (1 − w) * Apos EP The adaptive template is optimised for each PET image after the affine registration of the PET to AC C the template. After optimisation, the spatial normalisation to the template can be performed directly without the need for an MRI. The subsequent quantification steps are then identical to those performed as part of the standard SPM8 pipeline. 2.3. Level-1 centiloid replication analysis The first step is to check that we can replicate the results of the Level-1 centiloid calibration using the SPM8 standard pipeline. In order to improve clarity, the notations used in this manuscript might differ slightly from the ones employed in the original centiloid paper (7). ACCEPTED MANUSCRIPT The 50-70 minutes PiB PET images and corresponding MRI from the 34 YC, and 45 AD were downloaded from the GAAIN website. All images were then processed using the recommended SPM8 pipeline. SUVR values for each scan were computed using the standard cortex mask, and whole cerebellum as the reference region. Centiloid (CL) values were computed using the ∗ is the mean SUVR of the 34 YC, ∗ AD, and ∗ − ∗ ∗ ∗ M AN U Where − SC ∗ = 100 × RI PT following equation: is the mean SUVR of the 45 is the SUVR of an individual subject, with the asterisk denoting that the SUVR were computed at our site (using the standard SPM8 pipeline). This gave us the following equation: − 1.014 Equation1 1.073 ∗ TE D = 100 × We then performed a linear correlation of our centiloid values, to the ones reported in (7). The R2 EP was 0.999 and the intercept was -0.073 CL, both within the accepted range (R2>0.98 and intercept between −2 and 2 CL). For the remainder of this manuscript, the centiloid values computed using AC C the SPM8 pipeline at our site, as well as the centiloid equation defined in Eq 1, will be used in all calculations. 2.4. Level-2 Calibration of CapAIBL using PiB In order to perform centiloid quantification using CapAIBL, the SPM template used in the standard pipeline was spatially normalised to the adaptive template used in CapAIBL. The centiloid VOIs were then propagated accordingly to the adaptive template’s space. ACCEPTED MANUSCRIPT The 50-70 minutes PiB PET images from the 34 YC and 45 AD were spatially normalised to the adaptive template using CapAIBL, and their SUVR were computed using the centiloid VOIs. Calibrating a new method is a 2 steps process, where the PiB SUVR computed using the nonstandard method is first transformed into PiB-calculated SUVR or PiB-calcSUVR (as referred to in (7)), RI PT which are equivalent to the PiB SUVR computed using the standard SPM8 pipeline. These calc SUVR can then be transformed into centiloid using Eq 1. The first step is to compute the slope (PiBmCapAIBL) and Intercept (PiBbCapAIBL) of the SUVR SC i. PiB- '( ) * ~ M AN U computed using CapAIBL against those computed using the standard centiloid, so that , '( ) * × ∗ + . '( ) * Since the tracer is kept the same, but the quantification methods are different, the parameters (PiBmCapAIBL,PiBbCapAIBL) inform on the differences due to the use of CapAIBL. After PiB SUVRCapAIBL is converted into TE D checking that the R2 is above 0.7, PiB-calc SUVRCapAIBL using the standard method using the calculated slope and intercept, so that: PiB-calc '( ) * − , . '( ) * '( ) * SUVRCapAIBL is then converted into centiloid using Eq 1. AC C ii. '( ) * = EP /'0/ Both equations are combined to directly transform PiBSUVRCapAIBL into centiloid. 2.5. Level-2 Calibration of the 18F tracers using the standard method In this section, we will run the level-2 calibration of each 18 F tracer into centiloid using the standard SPM8 pipeline. For each tracer, its PET images and corresponding PiB PET and MR images were downloaded from the GAAIN website. Static images were available for all tracers except for the 18 ACCEPTED MANUSCRIPT F-Florebetapir-11C-PiB dataset where dynamic scans are the only images available. For this dataset, the individual frames were summed together. No frame-to-frame motion correction was applied, and no other pre-processing was done on the data. Basic demographics are provided in Table 1. RI PT Table 1. Demographics N Clinical group 11 79 34 YC, 45 AD 18 55 10 YC, 25 HC, 10 MCI, 7 AD, 3 FTD 18 35 10 YC, 6 HC, 9 MCI, 8 AD, 2 FTD 18 74 24 YC, 10 HC, 20 MCI, 20 AD 18 46 13 YC, 9 HC, 7 MCI, 3 Possible AD, 14 AD • MCI: mild cognitive impairment, FTD: frontotemporal dementia F-NAV4694 F-Florbetaben F-Flutemetamol AC C EP TE D F-Florbetapir M AN U C-PiB SC Tracer Figure 1. Overview of the level 2 centiloid calibration Summary ACCEPTED MANUSCRIPT The procedure to calibrate each tracer being identical, we will only describe it in general terms, and will provide the final equations in the results section. RI PT For a given tracer, its PET images and associated PiB PET images were processed using the standard SPM8 pipeline using the corresponding MR images. Neocortical SUVR for both the PiB and tracer’s PET images were computed using the centiloid VOIs. Computing centiloid values for a SC new tracer using the standard SPM8 pipeline is a 2 steps process (Figure 1.a), where the tracer’s transformed into centiloid using Eq 1. i. M AN U SUVR (TracerSUVRStd*) is first transformed into PiB-calculated SUVR (PiB-calcSUVRStd*), before being The first step is to compute the slope (TracermStd) and intercept (TracerbStd) of the tracer’s SUVR against those of the corresponding PiB, so that 12'/32 ~ 12'/32 , × ∗ + 12'/32 . TE D ∗ After checking that the R2 is above 0.7, TracerSUVRStd* is converted into PiB-calcSUVRStd* using the EP calculated slope and intercept, so that: ii. AC C /'0/ ∗ = 12'/32 − 12'/32 , ∗ 12'/32 . PiB-calc SUVRStd* is then converted into centiloid using Eq 1. Both equations are combined to directly transforms TracerSUVRStd* into centiloid. ACCEPTED MANUSCRIPT 2.6. Level-2 Calibration of the 18F tracers using CapAIBL In this section, we will run the level-2 calibration of each 18 F tracer into centiloid using CapAIBL. Similarly to the calibration using the standard method, the procedure to calibrate each tracer is identical, and therefore we will only describe it in general terms, and provide the final equations in RI PT the results section. For a given tracer, all the tracers’ PET images, as well as the associated PiB images were processed SC using CapAIBL. Neocortical SUVR for both the PiB and tracer’s images were computed. Computing centiloid values for a new tracer using a non-standard method is a 2 steps process (Figure 1.b), M AN U where the tracer’s SUVR (TracerSUVRCapAIBL) is first transformed into PiB-calculated SUVR (PiBcalc SUVRCapAIBL), before being transformed into centiloid using Eq 1. As specified in (7), the centiloid computed using the standard method using PiB should always be used as a reference. Therefore, the transform that converts (TracerSUVRCapAIBL) into PiB-calculated SUVR (PiB-calcSUVRCapAIBL), is i. TE D computed using the PiB SUVR from the standard method (PiBSUVRStd*) as the reference. The first step is to compute the slope (TracermCapAIBL) and intercept (TracerbCapAIBL) of the tracer’s '( ) * ~ 12'/32 , '( ) * × ∗ + 12'/32 . '( ) * AC C 12'/32 EP SUVR against those of the corresponding PiB, so that: After checking that the R2 is above 0.7, TracerSUVRCapAIBL is converted into PiB-CalcSUVRCapAIBL using the calculated slope and intercept, so that: /'0/ ii. PiB-Calc '( ) * = 12'/32 SUVR is converted into centiloid using Eq 1. '( ) * − 12'/32 , 12'/32 '( ) * . '( ) * ACCEPTED MANUSCRIPT Both equations are combined to directly transform TracerSUVRCapAIBL into centiloid. 2.7. Independent validation using images for the AIBL study RI PT In order to check the generalisability of the method outside the calibration dataset, the centiloid values computed using both CapAIBL and the standard method should be compared in an independent dataset. For each tracer, we used pairs of PET and MR images acquired as part of the SC Australian Imaging Biomarkers and Lifestyle (AIBL) study. The images were analysed using both the standard pipeline and CapAIBL and the resulting SUVR were converted into centiloid using M AN U their respective equations. The centiloid values were then compared in terms of agreement, bias and intra-class correlation (ICC). For the calculation of the ICC, we looked at the absolute agreement, using a two-way mixed-effects model, with single rater. TE D 3. Results 3.1. Level-2 Calibration of CapAIBL using PiB There was little bias between the PiB SUVR computed using CapAIBL and those computed using EP SPM, with the slope PiBmCapAIBL=0.991 and intercept PiBbCapAIBL=0.030. The R2 was 0.993, indicating AC C good agreement between both methods. The combined equation that converts PiBSUVRCapAIBL into CL was defined as: '( ) * = 100 × '( ) * − 1.034 1.063 The mean centiloid for the YC using CapAIBL was -0.04, and 100.3 for the AD subjects. The standard deviation in the YC was 4.40, compared to 4.34 when using the standard pipeline. ACCEPTED MANUSCRIPT A Scatter plot showing the centiloid values derived from CapAIBL against the ones derived from M AN U SC RI PT the standard pipeline is presented in Figure 2. Figure 2. Scatter plot of PiB centiloid values from the calibration dataset computed using CapAIBL TE D against the centiloid values computed using the standard method 3.2. Level-2 Calibration of CapAIBL using the 18F tracers Conversion into PiB-equivalent SUVR EP The equations used to convert the SUVR from each tracers into PiB-CalcSUVR are presented in AC C Table 2. The conversion equations computed using CapAIBL were similar to those produced using the standard method. For both quantification methods, there was a good agreement between the PiBSUVR and PiB-CalcSUVR, with a R2>0.89 for the standard method and R2>0.87 for CapAIBL, well above the suggested threshold of 0.70. For both method, NAV had the strongest agreement with PiB, followed by FLUTE, FBB and FBP. ACCEPTED MANUSCRIPT Table 2. Conversion equation for each 18F tracer SUVR into calculated-PiB SUVR using both SPM SPM /'0/ ; 18 F-Florbetaben 0.991 − 0.389 0.610 0.953 − 0.220 F-Flutemetamol 0.768 ; 18 F-Florbetapir /'0/ ϯ /'0/ @ @ − 0.046 0.980 1.008 ; '( ) * − 0.382 0.948 0.655 ;*>1? 0.962 '( ) * − 0.240 0.960 0.827 ; 0.893 '( ) * − 0.532 0.870 0.534 '( ) * EP '( ) * '( ) * 5ϯ TE D * − 0.518 0.517 '( ) * 7 8 + 0.073 1.090 ;*>1? 18 /'0/ 5∗ SC F-NAV4694 CapAIBL M AN U 7 8 18 RI PT and CapAIBL AC C Converting SUVR into centiloid The equations used to convert the SUVR from each tracers into centiloid are presented in Table Table 3. The scatter plots showing the corrected centiloid values from CapAIBL against those from the standard method are presented in Figure 3. The agreement between the centiloid computed using CapAIBL and those computed using the standard method was very high (R2>0.97) for all tracers. M AN U SC RI PT ACCEPTED MANUSCRIPT TE D Figure 3. Scatter plot of the corrected centiloid values computed for each 18F tracer using CapAIBL, against the corresponding centiloid values computed for the same tracers using the standard AC C EP method. ACCEPTED MANUSCRIPT Table 3. Equations to convert SUVR into centiloid. CL Equation (SPM) 100 × 7 8 18 F-NAV4694 100 × ; 18 F-Florbetaben 100 × F-Flutemetamol 100 × − 1.033 1.170 100 × − 1.007 0.654 100 × F-Florbetapir '( ) * '( ) * PiB 5ϯ NA 0.993 − 1.068 0.990 0.988 1.082 ; '( ) * 0.702 ;*>1? − 1.046 '( ) * 100 × 0.887 ; − 1.078 '( ) * − 1.073 0.990 0.995 0.974 0.988 0.986 0.993 0.573 M AN U 100 × − 1.034 5∗ 1.063 7 8 '( ) * '( ) * TE D @ − 1.042 0.554 100 × @ 12'/32 − 0.998 0.824 ; 18 ϯ 100 × ;*>1? 18 * 12'/32 − 1.014 1.073 Tracer RI PT C-PiB SC 11 CL Equation (CapAIBL) 3.3. Validation on independent datasets from AIBL 11 C- EP From the AIBL study, pairs of PET and MR images were selected for each tracer: 1000 pairs of PiB/MR, 79 of 18F-NAV4694/MR, 119 of 18F-Florbetaben/MR, 446 of 18F-Flutemetamol/MR and 278 of F-Florbetapir/MR. Relevant demographics for each tracer are presented in Supplemental Table 1. AC C 18 Scatter plot of the centiloid values computed using CapAIBL and the standard method are presented in Figure 4. The centiloid values derived from PiB using CapAIBL were in good agreement with those computed using the standard method (R2=0.990) and the slope of the regression shows a small underestimation of less than 0.8%. With NAV, while the agreement was good (R2=0.986), the slope of the regression showed an overestimation of 8.1%. With FBB, the agreement was also good (R2=0.982), but the slope of the regression showed an underestimation of 11.4%. With FLUTE, the ACCEPTED MANUSCRIPT agreement was also good (R2=0.977), and the slope of the regression showed very little bias, with a small overestimation of just 0.8%. With FBP, the agreement was a bit lower (R2=0.964), but the slope of the regression showed very little bias, with an overestimation of just 1.3%. For all tracers, the agreement defined by ICC was excellent (ICC>0.98). While the centiloid scale is not designed for RI PT image visualisation, all the images can now be presented using the same colour scale. Example images for each tracer from the AIBL study are presented in Supplemental Figure 1 for centiloid AC C EP TE D M AN U SC values of 0, 25, 50, 75 and 100. AC C EP TE D M AN U SC RI PT ACCEPTED MANUSCRIPT Figure 4. Scatter plot of the corrected centiloid values computed on the AIBL dataset for each 18 F tracer using CapAIBL, against the corresponding centiloid values computed for the same tracers using the standard method. ACCEPTED MANUSCRIPT 4. Discussion This study showed that the PET-only quantification method, CapAIBL, can be used to perform centiloid quantification using the most commonly used Aβ PET tracers. CapAIBL generated SUVR values that were in very good agreement with those generated using the standard pipeline for all RI PT tracers on the calibration data. We have also illustrated how a new method can be calibrated to generate centiloid values. SC On an independent dataset from the AIBL study, the centiloid values computed for PiB, FLUTE and FBP using the standard pipeline and CapAIBL were in very good agreement, with very little bias, M AN U which would indicate that the method generalises well and can be used in other centres. With NAV and FBB however, while the R2 was high, there was significant bias between the methods. There are several possible causes for this. TE D Size of dataset. We had access to much a smaller dataset for NAV and FBB compared to PiB, FLUTE and FBP which could make the estimate of the regression parameters less accurate. Moreover, for these 2 tracers, there was a greater proportion of subjects with very low CL value (54% of subjects EP imaged with FBB and 57% of subjects imaged with NAV have CL<10, compared to 43% and 49% for FLUTE and FBP). Therefore, a few positive outliers could influence the estimation of the regression AC C parameters. While only a limited dataset was available at the time of this study, we will be able to further refine the estimate of the regression parameters as more images become available. Scanner effect. The use of different scanners could bias the SUVR estimates with CapAIBL, as different point spread functions, reconstruction algorithms, and acquisition parameters can alter the appearance of the PET images (15,16). This could in turn introduce a bias during the estimation of the adaptive atlas and the spatial normalisation to that atlas. To further examine this, we compared the scanners used for the acquisition of the calibration scans and the AIBL scans for NAV and FBB. All the ACCEPTED MANUSCRIPT calibration scans for NAV were acquired on a Philips Allegro, and in AIBL, 55 of the scans were acquired on the same scanner, while 24 were acquired on a Philips Gemini. In addition to different scanners being used, different reconstruction algorithm were employed, with a 3D row action maximum likelihood algorithm (3D RAMLA) used on the Allegro, and line of response RAMLA (LOR RI PT RAMLA) on the Gemini. Figure 5 presents the scatter plot of the Centiloid values computed using CapAIBL against the ones computed using the standard method segregated by scanner model. By restricting the AIBL dataset to the scans acquired on the same scanner as the calibration dataset SC (Allegro scanner), the error is reduced to a slight overestimation of 2.6%. With the scans acquired on the Gemini scanner, the error is now much larger, with an overestimation of 14.5%. This would M AN U indicate that the SUVR computed using CapAIBL are scanner dependant, and that by using the same scanner on both the calibration dataset and the independent dataset, very accurate quantification can be obtained. We could not replicate a similar plot with FBB, as most of the calibration scans (29/35) were acquired on the Gemini scanner, while most of the AIBL scans (112 /119) were acquired TE D on the Allegro scanner. However, the differences observed in NAV, where the CL computed using CapAIBL was about 12% higher in scans acquired on the Gemini scanner compare to those acquired EP on the Allegro scanner, is comparable to the differences seen in FBB, where the AIBL scans acquired using the Allegro had CL 11.4% lower than what they should be given the calibration data that were AC C acquired on the Gemini scanner. From the available data, we can estimate that using CapAIBL, there is a 12% difference in CL measures between the Allegro and Gemini scanners in both NAV and FBB, with CL computed on Gemini scans being higher than the ones from Allegro, and that the difference in scanner can explain most of the differences observed between the centiloid values computed using CapAIBL and the standard method in the AIBL dataset. While here, we assumed that the standard method provided the ground truth, and only focused on the differences between CapAIBL and the standard method given different scanners, it is worth noting that some of these scanner ACCEPTED MANUSCRIPT specific differences could also introduce biases when using the standard pipeline, as previous studies have shown inter-scanner image variability in multi-centre studies, requiring scanner specific M AN U SC RI PT calibrations (15,16). Figure 5. Scatter plot of the corrected centiloid values computed on the AIBL dataset for NAV for each scanner using CapAIBL, against those computed using the standard method. TE D Tracer specific differences. Lastly, different tracers might be more susceptible to exhibit large differences when different scanner/reconstruction method are used, which would in turn produce EP differences in the CL values computed using CapAIBL. With the current data at hand, it is hard to pin point to the exact origin of some of the differences AC C observed between the CL values computed using CapAIBL and the standard method. There is however a very strong indication that the use of different scanners and reconstruction algorithm could be responsible for some of these differences. The exact source of these differences, and how much of an effect they have on centiloid values computed using different methods and different tracers needs to be further examined. In the mean-time, since the bias appears to be a linear transform, this could be accounted for using a scanner specific calibration transform. While this cannot account for potential differences in the CL values computed using the standard method with ACCEPTED MANUSCRIPT different scanner, it can ensure that CL values computed using CapAIBL are consistent with the ones computed using the standard method. RI PT 5. Conclusions This is the first study showing the calibration of SUVR values produced using a non-standard PET quantification method to the centiloid scale. Using CapAIBL, we showed that reliable centiloid SC estimates could be obtained using a PET-only quantification method, which could facilitate its M AN U adoption in the clinic, as the centiloid values could be produced directly on the PET scanner. The bias observed in NAV and Florbetaben images from the AIBL study indicates that a scanner specific linear correction might be required. Further work is required to better understand and characterise the variability induced by the use of different scanners and reconstruction algorithms in both CapAIBL Disclosures TE D and the standard quantification pipeline. EP Vincent Doré, Olivier Salvado, Jurgen Fripp, Chris Rowe, Victor Villemagne are inventors on patent US9361686B2 that describes some aspects of the software CapAIBL. Chris Rowe has received AC C research grants from Piramal Imaging, GE Healthcare, Cerveau, Astra Zeneca, Biogen. Victor Villemagne is and has been a consultant or paid speaker at sponsored conference sessions for Eli Lilly, Piramal Imaging, GE Healthcare, Abbvie, Lundbeck, Shanghai Green Valley Pharmaceutical Co Ltd, and Hoffmann La Roche. Bibliography 1. Rowe CC, Villemagne VL, Brain Amyloid Imaging. J Nucl Med. 2011;52:1733–1740. ACCEPTED MANUSCRIPT 2. Carrillo MC, Rowe CC, Szoeke C, et al. Research and standardization in Alzheimer’s trials: Reaching international consensus. Alzheimers Dement. 2013;9:160–168. 3. Clark CM, Pontecorvo MJ, Beach TG, et al. Cerebral PET with Florbetapir compared with neuropathology at autopsy for detection of neuritic amyloid-β plaques: a prospective cohort study. Lancet Neurol. RI PT 2012;11:669–678. 4. Rowe CC, Ng S, Ackermann U, et al. Imaging beta-amyloid burden in aging and dementia. Neurology SC 2007;68:1718–1725. 5. Rowe CC, Ackerman U, Browne W, et al. Imaging of amyloid beta in Alzheimer’s disease with 18F-BAY94- M AN U 9172, a novel PET tracer: proof of mechanism. Lancet Neurol. 2008;7:129–135. 6. Thurfjell L, Lundqvist R, Buckley C, et al. Automated quantification of [18F]Flutemetamol data Comparison with standard of truth based on histopathology. J Nucl Med. 2013;54:302–302. 7. Klunk WE, Koeppe RA, Price JC, et al. The Centiloid Project: Standardizing quantitative amyloid plaque TE D estimation by PET. Alzheimers Dement. 2015;11:1–15. 8. Rowe CC, Jones G, Doré V, et al. Standardized Expression of 18F-NAV4694 and 11C-PiB β-Amyloid PET EP Results with the Centiloid Scale. J Nucl Med. 2016;57:1233–1237. AC C 9. Rowe CC, Doré V, Jones G, et al. 18F-Florbetaben PET beta-amyloid binding expressed in Centiloids. Eur J Nucl Med Mol Imaging. 2017;44:2053–2059. 10. Battle MR, Buckley CJ, Smith A, et al. Utility of PMod image quantification software for processing [11C]PIB and [18F]FLUTEMETAMOL images for SUVR quantification on the centiloid scale [abstract]. Alzheimers Dement. 2016;12(supp 1):126. 11. Navitsk, M, Joshi AD, Devous MD, et al. Conversion of amyloid quantification with Florbetapir SUVR to the centiloid scale [abstract]. Alzheimers Dement. 2016;12:25–26. ACCEPTED MANUSCRIPT 12. Bourgeat P, Villemagne VL, Dore V, et al. Comparison of MR-less PiB SUVR quantification methods. Neurobiol Aging. 2015;36(Suppl 1):159-166. 13. Bourgeat P, Dore V, Fripp J, et al. Computational analysis of PET by AIBL CapAIBL: a cloud-based processing pipeline for the quantification of PET images [abstract]. J Nucl Med. 2015;56(suppl 3):149 RI PT 14. Edison P, Carter SF, Rinne JO, et al. Comparison of MRI based and PET template based approaches in the quantitative analysis of amyloid imaging with PIB-PET. NeuroImage. 2013;70;423–433. SC 15. Habert M-O, Marie S, Bertin H, et al. Optimization of brain PET imaging for a multicentre trial: the French CATI experience. EJNMMI Phys. 2016;3:6 AC C EP TE D NeuroImage. 2009;46;154–159. M AN U 16. Joshi A, Koeppe RA, and Fessler JA. Reducing between scanner differences in multi-center PET studies.