Computational Analysis of PET by AIBL (CapAIBL): A cloud-based processing pipeline for the quantification of PET images

Objectives Evaluate a recently developed cloud based analysis method for Positron Emission Tomography (PET) on a database of C11 and F18 β-Amyloid (Aβ) tracers as well as F18-FDG.

Methods The Computational Analysis of PET by AIBL (CapAIBL) is a publically available cloud based platform (https://capaibl-milxcloud.csiro.au) where PET images are spatially normalised to a standard template using an adaptive atlas approach [1], SUVR normalised and quantified. Four hundred and fifty four participants underwent MRI and PET scans with 18F-Flutemetamol (N=180), 11C-PiB (N=381), 18F-Florbetapir (N=171), 18F-Florbetaben (N=148),18F-NAV4694 (N=47) and 18F-FDG (N=34). Each PET image was analysed using CapAIBL. The SUVR normalisation was performed using each tracer’s reference region (Cerebellum GM for 11C-PIB, 18F-Florbetaben, 18F-NAV4694 and 18F-FDG, Pons for 18F-Flutemetamol and Whole Cerebellum for 18F-Florbetapir). For validation purposes, the images were also quantified using their corresponding MR. The error in neocortical SUVR between CapAIBL PET-only approach and the MR-based quantification was assessed using the coefficient of determination (R²) and mean absolute percentage error (MAPE).

Results The error in neocortical SUVR quantification and the coefficient of determination are reported in Table 1. The mean error in neocortical SUVR is lower than 5% and is comparable across tracers.

Conclusions As the use of PET Aβ tracer becomes more prevalent, there is going to be a greater need for standardised methods to analyse and quantify these images. CapAIBL can accurately quantify Aβ PET images without MR, with a similar degree of accuracy across tracers.