Abstract
1792
Objectives Several software packages have been developed to compute the left ventricular ejection fraction (EF). All approaches implement some prior allowing the user to start calculation with an initial target. In this study, we compare the EF obtained in small animals from three semi-automatic programs: MI-Heart (MI), Quantitative PET (QPET), and Corridor4DM (4DM), as well as manual tracing.
Methods The dataset contained images of 30 rats and 33 mice with various health conditions: 2 infarcted, 14 aortic regurgitation, 26 heart failure, and 21 healthy controls. Images were recorded with three different radiotracers (18F-FDG, 18F-FTHA and 11C-Acetate) on LabPET scanners. Two observers processed the images up to three times with each software, except QPET used only once each since images were already reoriented. Manual tracing was performed to provide the EF reference standard (STD). Bland-Altman plots and correlation were used to demonstrate the difference and limits of agreement between observers.
Results MI-Heart had an average bias relative to STD of -9.5% for rats and -5.3% for mice, while QPET and 4DM had a bias of -19.1% and -19.6% for rats and +9.4% and +10.5% for mice, respectively. A good correlation of individual EF data (r > 0.7) was observed between software for rats as well as for mice. Test-retest repeatability was similar between observers, assessed by the coefficients of repeatability (CR): 7.6% MI, 6.1% 4DM and 6.2% QPET for rat and 13.6% MI, 11.9% 4DM and 12.5% QPET for mice.
Conclusions All three software yielded less variable results with less bias relative to STD on mice than on rats. Overall, the dedicated MI-Heart program performed as well as adapted clinical software to evaluate the ejection fraction in small animals.