Gated myocardial perfusion SPECT underestimates left ventricular volumes and shows high variability compared to cardiac magnetic resonance imaging -- a comparison of four different commercial automated software packages

N = 100 patients with known or suspected coronary artery disease were examined at rest with 99 mTc-tetrofosmin gated MPS and cardiac MR imaging. Left ventricular end-diastolic volume (EDV), end-systolic volume (ESV), stroke volume (SV) and ejection fraction (EF) were obtained by analysing gated MPS data with four different programs: Quantitative Gated SPECT (QGS), GE MyoMetrix, Emory Cardiac Toolbox (ECTb) and Exini heart.All programs showed a mean bias compared to MR imaging of approximately -30% for EDV (-22 to -34%, p < 0.001 for all), ESV (-12 to -37%, p < 0.001 for ECTb, p < 0.05 for Exini, p = ns for QGS and MyoMetrix) and SV (-21 to -41%, p < 0.001 for all). Mean bias ± 2 SD for EF (% of EF) was -9 ± 27% (p < 0.01), 6 ± 29% (p = ns), 15 ± 27% (p < 0.001) and 0 ± 28% (p = ns) for QGS, ECTb, MyoMetrix, and Exini, respectively.Gated MPS, systematically underestimates left ventricular volumes by approximately 30% and shows a high variability, especially for ESV. For EF, accuracy was better, with a mean bias between -15 and 6% of EF. It may be of value to take this into consideration when determining absolute values of LV volumes and EF in a clinical setting.Gated myocardial perfusion single photon emission computed tomography (MPS) has been shown to provide diagnostically [1] and prognostically [2] important clinical information. A number of different software programs for determining left ventricular (LV) volumes by MPS have been developed. These different programs employ varying algorithmic approaches to quantify LV volumes. Such programs include Quantitative Gated SPECT (QGS) [3], Emory Cardiac Toolbox (ECTb) [4], MyoMetrix [5] and Exini heart [6]. In summary, the four programs use different automated algorithms which all go through three roughly similar steps in order to segment the LV. The first step is to approximate the location of the LV in the image. Secondly, the LV myocardial midmural centerline is detected within this location. Thirdly, the endocardia