Abstract
3188
Introduction: Dynamic cardiac positron emission tomography (PET) imaging extends functional imaging by incorporating information from multiple time frames, as opposed to the clinically used “static” rest/stress perfusion imaging. The utility of dynamic PET is well established in the research literature. Yet, many practical challenges render the technique unsuitable in clinical scenarios. One of the challenges is a relatively long scan time. In our previous work, we have shown that a matrix factorization technique can be used on input dynamic images to generate spatial regions of interest (ROIs) as well as the factor curves representing tracer dynamics in those ROIs. We report here preliminary results of applying matrix factorization on varying time windows of dynamic cardiac PET data to reveal how shorter scans affect the factorization results.
Methods: Our novel matrix factorization algorithm, called clustering-initiated factor analysis (CIFA), was applied to 13N-ammonia cardiac PET data. CIFA factorizes the dynamic images into three ROIs (left ventricular myocardium, left and right ventricular chambers) and their respective tracer-activities as three independent factor curves. The novelty of our algorithm is data-driven initialization called R-clustering that we invented to identify representative time-activities in the input dynamic images. We applied the CIFA algorithm on reduced time windows (8 of them) to study the quality of output. The resultant 3D coefficient images and factor curves were further analysed, and the myocardium coefficients were identified using the shape of the respective factor curves.
Results: The Pearson's correlation coefficient (PCC) was used to compare the shortened time windows' 3D coefficient images against that from the full-length dynamic imaging (600 seconds), considered as the ground truth. As the imaging window shortens, the PCC begins to decrease, as expected, prior to a large drop. This quality degradation can also be seen on the shape of the factor curves. The time window at 300 seconds resulted in a PCC of 0.98.
Conclusions: We demonstrated the ability of a shorter time window (300 seconds) retaining a high Pearson's correlation when compared to the ground truth, full-length time window (600 seconds). We expect our result to open an opportunity for developing robust reduced time dynamic cardiac PET imaging protocol.