Using PET to determine the ath-erosclerotic burden in the cardio-vascular system

Research output: ThesisPh.D. thesis

86 Downloads (Pure)


Introduction: Ischemic heart disease and stroke are the world’s number one and two killers. The underlying cause is usually atherosclerosis, which may stay asymptomatic for years and is usually diagnosed late in the course due to a complication or during a health check using common types of imaging such as ultrasound with Doppler, computed tomography (CT) or magnetic resonance imaging, all of which can also determine if a stenosis is present. Recent reports suggest that very early changes in the artery wall can be detected and measured by positron emission tomography (PET) imaging with tracers 18F-fluorodeoxyglucose (FDG) or 18F-sodium fluoride (NaF). In this PhD project, we initially wanted to study atherosclerosis in the carotids by PET imaging, but we soon changed our focus to also include the heart and aorta and the use of AI to segment the targeted structures in order to elucidate if this could make segmentation and quantification faster and more reliable. Thus, the project ended comprising four studies published in four articles. Article I was a systematic review on PET imaging of carotid atherosclerosis, emphasizing clinical usefulness and relations to conventional imaging modalities. Article II was a study of 2-year changes in carotid and aortic NaF uptake in healthy individuals and angina patients using conventional manual segmentation. Article III was an attempt to establish and test an automated AI-based method for fast segmentation of the heart in NaF-PET/CT scans, while Article IV was an attempt to do the same in the aorta.

Methods: In the systematic review, articles on carotid artery PET imaging with different radiotracers were searched in several databases. Duplicates, editorials, case stories, studies regarding feasibility or reproducibility of PET imaging, and studies on patients with end-stage diseases or receiving immunosuppressive medication were omitted. All eligible articles were reviewed by one observer. In the cohort study using manual segmentation only, 29 healthy subjects and 20 angina pectoris patients underwent NaF-PET/CT twice two years apart. The arch, thoracic, and abdominal aorta and the carotids were manually segmented. NaF uptake was expressed as the maximum, mean and total standardized uptake values without and with partial volume correction (SUVmax, SUVmean, SUVtotal and cSUVmean, cSUVtotal). Subsequently, a convolutional neural network (CNN) based method was developed to identify and segment the heart and the aorta in three mentioned parts. The CNN model was trained in NaF- PET/CT scans of other patients and tested in the same 49 subjects as above by comparison with data obtained by manual segmentation. Bland-Altman limits of agreement were used to compare derived parameters. Furthermore, the reproducibility of the manual method was examined by repeated segmentation in 25 randomly selected scans.

Results: In the systematic review, it was shown that patients with symptomatic carotid atherosclerosis have higher FDG uptake than patients with asymptomatic carotid atherosclerosis. There was a strong correlation between microcalcification and NaF uptake in symptomatic patients in histopathological assessment, but calcification had a negative correlation with uptake of FDG. In the manual cohort study, NaF uptake was insignificantly higher in the angina group at both time points, with less uptake in the healthy group and slightly higher uptake in the angina group after two years. NaF uptake at baseline could not predict a change in CT calcification after 2 years. NaF uptake correlated positively with age in all parts of the aorta. CT scan did not indicate any change in density of major arteries after 2 years of follow-up. In the final part of the project, CNN derived heart segmentation measures were 0% to 4% higher than by the manual method and 0% to 17% lower than with manual aortic segmentation. However, with CNN-based and manual method the SUVmean values in both heart and aorta were almost identical. Cardiac and aortic CNN-based segmentation method was much faster than the manual approach, which had a maximal 0.5% and 6% variation at repeated segmentation of the heart and the aorta, respectively, compared to a 100 % inborn CNN reproducibility.

Conclusion: PET imaging is a newly introduced modality for imaging of atherosclerosis, which is a slow and variable process in healthy individuals and patients with angina pectoris, albeit with a tendency of slightly higher NaF uptake in angina patients. Although current technical difficulties such as time-taking image analysis exist, the AI-based models could present values for Volume, SUVmean, SUVmax, and SUVtotal similar to the manually obtained ones. These AI-based models are observer-independent, highly reproducible and very fast alternative alternatives for slow manual segmentation. With further training, the AI-based approach may become the standard for assessing patients with suspected or known atherosclerosis.
Original languageEnglish
Awarding Institution
  • University of Southern Denmark
  • Høilund-Carlsen, Poul, Principal supervisor
  • Alavi, Abass, Co-supervisor, External person
  • Olsen, Michael Hecht, Co-supervisor
  • Gerke, Oke, Co-supervisor
Publication statusPublished - 31. May 2022


Dive into the research topics of 'Using PET to determine the ath-erosclerotic burden in the cardio-vascular system'. Together they form a unique fingerprint.

Cite this