Validation and reproducibility of mapping positron emission tomography uptake in the aortic wall and thrombus

Abstract

INTRODUCTION: Aortic aneurysms are prevalent diseases of the aorta, with limited knowledge about their aetiology, progression, and risk, and no effective pharmacological treatment. 18F-fluorodesoxyglucose Positron Emission Tomography (18F-FDG PET) provides molecular level information of glucose activity, serving as potential analogue of vascular inflammation, which may be relevant to improve knowledge and clinical management of aortic aneurysms. Nonetheless, current clinical assessment of 18F-FDG uptake presents several limitations. PURPOSE: The aim of this study was to develop, test and validate an innovative aortic wall uptake mapping. METHODS: PET/magnetic resonance (MR) data from 72 patients were acquired. The aortic lumen was segmented, and 9 anatomical landmarks were identified to create a standardised aortic lumen discretization including up to 80 patches. Black-blood MR images were used to guide thrombus segmentation if present, which was merged with lumen segmentation to generate a thoraco-abdominal aortic wall mask. This mask was then expanded by 1 to 5 mm in the inward and outward directions, resulting in an aortic wall volume of interest. Median and percentile 95th standard uptake value (SUV) was calculated on each aortic patch. A second observer performed the same analysis on 23 randomly selected patients for inter-observer reproducibility assessment. Validation was performed by comparisons with regional SUV measured by a nuclear medicine expert. RESULTS: The technique was highly feasible, permitting the analysis of all 72 patients and in the 99.6% of aortic wall patches. The image analysis workflow was highly reproducible, resulting in Dice scores of 0.90 [0.57, 0.91] and 0.85 [0.78, 0.88] for aortic and thrombus segmentations, respectively, and in excellent co-localization for anatomical references (5.74 [3.62, 8.73] mm). The inter-observer reproducibility of aortic wall SUV mapping was excellent (ICC between 0.924 and 0.945), with limited differences with respect to aortic wall thickness and similar performances for SUV quantification in the thrombus. The validation of regional SUV values showed good agreement, with limited impact of aortic wall thickness values. A balance between reproducibility and accuracy was obtained with a volume of interest of 6 mm thickness. CONCLUSIONS: An image analysis implementation based on multi-modality PET/MR data provides reproducible and accurate quantitative aortic wall 18F-FDG uptake maps. This approach may enable exploring local factors related to an inflammatory vascular state, with possible repercussions in clinical practice.