TY - JOUR T1 - <strong>Centiloid harmonisation across PET cameras without using paired data</strong> JF - Journal of Nuclear Medicine JO - J Nucl Med SP - 3176 LP - 3176 VL - 63 IS - supplement 2 AU - Shenpeng Li AU - Pierrick Bourgeat AU - Leo Lebrat AU - Graeme O'Keefe AU - Jurgen Fripp AU - Victor Villemagne AU - Christopher Rowe AU - Vincent Dore Y1 - 2022/06/01 UR - http://jnm.snmjournals.org/content/63/supplement_2/3176.abstract N2 - 3176 Introduction: The Centiloid (CL) scale allows quantification of amyloid-b (Aβ) accumulation across multiple PET tracers for the diagnosis of the Alzheimer’s disease (AD). However, differences in PET cameras especially from different manufacturers or using different technology yield slightly different CL which results in significant quantification bias when merging cohorts acquired on different cameras or in the longitudinal studies when change of camera occurs. The CL difference may be modelled by paired scans, where the same subjects are scanned on two different cameras at the same time. However, paired scans are often not available. In this study, we propose a harmonisation method, based on the non-negative matrix factorisation (NMF), to reduce the quantification bias between different cameras without using paired scans. We compared the harmonisation for two pairs of PET cameras, Philips Gemini TF64 versus Philips Allegro and Philips Gemini TF64 versus Siemens Biograph mCT.Methods: DataThree hundred and forty-eight 18F-NAV4694 scans from the Australian Imaging Biomarkers and Lifestyle (AIBL) study acquired on three PET cameras Siemens Biograph mCT, Philips Allegro and Philips Gemini TF64 (N=116 each), were used to train the model. Participants who switched camera between two consecutive timepoints (TP) were selected for the validation dataset. Seventeen participants were imaged on the Siemens mCT and Phillips Gemini TF64 (2.63±0.51 years scan interval), and fifty-two participants were imaged on the Philips Gemini TF64 and Phillips Allegro (2.73±0.73 years). One hundred participants who had two scans (2.26±0.82 years) on Gemini TF64 were used to define a reference rate of change (CL/year). All PET images were processed using CapAIBL, a PET-only quantification pipeline to calculate the CL and provide the processed CL images in standard space for the PET camera harmonisation. Harmonisation:Using a modified NMF approach, the images from two different cameras were jointly decomposed into three components: a biological specific binding component which is identical for both cameras, a camera specific component &lt;m:omath&gt;H&lt;/m:omath&gt;&lt;m:omath&gt;C&lt;/m:omath&gt; which represents the unique features of the camera, and a non-specific binding component D which represents a very negative subject imaged by that camera. The CL values are then harmonised by transferring the camera unique components &lt;m:omath&gt;C&lt;/m:omath&gt; and &lt;m:omath&gt;D &lt;/m:omath&gt;from one camera to the other for each PET image. Figure 1 illustrates the harmonisation workflow.Results: Figure 2 compares the longitudinal CL over time between Gemini TF64 (TP1) and mCT (TP2). The harmonisation leads to a reduction in the CL/year, especially for the negative (CL&lt;20) and very positive (CL&gt;100) subjects whose rate of change is expected to be close to 0 CL/year. Figure 3 compares the curves of CL rate of change against baseline CL for mCT to Gemini compared to the reference (black). The harmonised fitting (blue) is closer to the reference curve than the one without harmonisation (red) in the low to middle CL range. The fitting in high CL range (CL&gt;80) deviates from the reference curve because of the lack of subjects in this range. Figures 4 and 5 present the harmonisation results between the Allegro (TP1) and Gemini (TP2). The harmonised fitting in Figure 5 is nearly parallel to the reference curve for the entire CL range. In contrast, the fitting on the unharmonized data shows negative rate of change for the very negative subjects and much higher CL change rate for the very positive subjects, which are strongly different compared to the reference curve.Conclusions: We propose a practical and economical CL harmonisation for different PET cameras without using paired data. The NMF based method shows promising results for scanner harmonisation, though a larger sample size in the high CL range (CL&gt;80) is needed in future studies for the validation between Gemini TF64 and Siemens mCT. ER -