Alignment of Cardiac Structures in [15O]H2O Perfusion PET Studies

Abstract

Positron emission tomography (PET) with [15O]H2O provides a method to measure quantitatively the myocardial perfusion in vivo. The quantification enables an objective and accurate way to compare studies acquired on different time occasions or under different conditions. The challenge in [15O]H2O cardiac perfusion PET studies is the low contrast between cardiac structures in the images, because the tracer is freely diffusible. Furthermore, the contraction motion of the heart and the movement of the heart in the thorax, due to respiration, make the quantitative analysis even more difficult. In addition, the patient may move during the acquisition or the position of the patient may change between the studies. For these reasons, the goal of the thesis has been to develop and evaluate image analysis methods to improve the usability of [15O]H2O myocardial perfusion PET studies for drug development and clinical diagnostics. In this thesis, a novel systemic process to align the studies to the same position is presented. Thanks to the alignment, the same regions of interest (ROIs) can be used in the analyses of the studies, ensuring that the comparison is carried out in equivalent myocardial segments. This increases the repeatability and accuracy of the quantification. Because the heart position is not fixed in the thorax, the alignment should be carried out using cardiac structures instead of external landmarks or other thoracic structures. In this thesis, the independent component analysis (ICA) method was evaluated for the separation of the cardiac structures. For the alignment of the studies, an approach based on mutual information (MI) was developed. The difficulty in developing and improving the analysis of medical images is that reference values, i.e. golden standards, are not available, on which to base the comparison. In this thesis, one of the aims was to construct realistic simulated data, to be able to evaluate the developed approach. In the constructed simulated data, knowledge of the heart physiology, anatomy and function was combined in the measurement of the myocardial perfusion with PET and O-15- labeled water. With the simulated data, it was possible to quantitatively evaluate ICA for the separation of the cardiac structures, as well the developed approach to align the studies. The cardiac structures were successfully separated with ICA under different perfusion levels, and with cardiac and respiratory motion, even with real human data. The results proved that the cardiac structures from different studies can reliably be extracted for the alignment. The quantitative evaluation of the developed approach proved that by aligning the studies, the same ROIs can be used for comparative analysis of the studies. Furthermore, the alignment of the two studies could be carried out without the use of information from the myocardium, and thus the method is not affected by possible abnormalities in the myocardial tissue. The developed approach can be used, in practice, with real human studies to facilitate analysis of [15O]H2O myocardial perfusion PET studies.