J Nucl Med. 2007; 48 (Supplement 2):408P
Instrumentation & Data Analysis: Data Analysis & Management
Data Analysis & Management Posters |
Anatomical accuracy & variability in factor analysis of dynamic structures (FADS) with cardiac 18FDG PET imaging
Ran Klein1,
M'hamed Bentourkia4,
Andy Adler3,
Jean DaSilva1,
Richard Wassenaar1,
Robert Beanlands1 and
Robert deKemp1
1 Cardiac PET, Ottawa Heart Institute, Ottawa, Ontario, Canada; ;
2 SITE, University of Ottawa, Ottawa, Ontario, Canada; ;
3 CSE, Carleton University, Ottawa, Ontario, Canada; ;
4 Nuclear Medicine and Radiobiology, University of Sherbrooke, Sherbrooke, Quebec, Canada
1712
Objectives: Factor analysis has been explored for the past two decades as a means of decomposing dynamic medical images into their structural components. However, few have validated the anatomic accuracy of their results. In this work, we demonstrate a validation technique within the context of cardiac PET. Methods: One dog was imaged on six separate occasions (14 days apart) using a static 11CO scan followed by a dynamic 18FDG scan after bolus injection. 18FDG scans were cropped semi-automatically to remove adjacent organs and then decomposed using FADS into 2-4 structures. Ideally, we expect to resolve LV blood, RV blood, myocardium, and liver structures. The structures associated with blood-pool anatomy were summed to produce a total-blood-factor image, and correlated with the 11CO images. CO binds strongly to hemoglobin, so the resulting image of the blood-pool should correspond to the total-blood-factor. Results: Overall, the best correlation was obtained with 3 factors. In most cases good correlation (R2
0.75) was obtained regardless of number of factors, and very good correlation (R20.9) was obtained in 2/6 studies. Moderate-to-poor correlations were observed in 2 studies where the early blood-pool frames were missing. The correlations varied significantly (ANOVA) between studies (P<0.001), but not with the number of factors (P=0.92), suggesting more dependence on the dynamic imaging protocol, as opposed to number of factors selected. Conclusions: Automated FADS can produce accurate blood-factor images from dynamic 18FDG imaging. However, more standardized imaging protocols may be needed to obtain reproducible results.
| |
Cross-correlation of blood-factor and CO image
|
|
