Validation of QGS and 4D-MSPECT for Quantification of Left Ventricular Volumes and Ejection Fraction from Gated ¹⁸F-FDG PET: Comparison with Cardiac MRI

Patients

We examined 44 patients (37 male, 7 female; mean age, 65 ± 8 y; range, 48–81 y) who had severe CAD and were scheduled for routine ¹⁸F-FDG PET to assess myocardial viability. Of these, 36 had a history of at least 1 previous myocardial infarction, 15 had 1-vessel disease, 9 had 2-vessel disease, and 20 had 3-vessel disease. Diabetes mellitus had previously been diagnosed in 13 patients. The mean fasting glucose level was 131 ± 61 mg/dL (range, 81–430 mg/dL). All patients gave informed consent for cMRI. The local ethics committee approved the study.

PET Methodology

PET scans were obtained on an ECAT EXACT 922/47 scanner (Siemens-CTI). All patients received 250 mg of acipimox 2 h before administration of ¹⁸F-FDG. One hour before injection of ¹⁸F-FDG, nondiabetic patients received an oral glucose load of 50 g to stimulate endogenous insulin secretion. Instead of receiving glucose, diabetics received insulin (3–30 IU) intravenously 5–10 min before administration of ¹⁸F-FDG. Gated acquisition was done 60 min after intravenous administration of 283 ± 38 MBq of ¹⁸F-FDG, dividing the cardiac cycle into 8 equal intervals. The acquisition time was 30 min for emission (2-dimensional mode) and 15 min for transmission (⁶⁸Ge/⁶⁸Ga rod sources). After attenuation and scatter correction of the gated ¹⁸F-FDG PET sinograms, reconstruction using filtered backprojection in a 128 × 128 matrix was done for all gates (Hanning filter; cutoff frequency, 0.4 Nyquist). A zoom factor of 1.525 was applied to obtain transverse images with an isotropic voxel size of 3.38 mm per side. The isotropic datasets were transferred to a Siemens e.soft workstation (Siemens Gammasonics Inc.), where they were reoriented on the transverse planes, with reorientation parallel first to the septum and then to the inferior wall. The resulting reoriented datasets were stored for analysis.

Quantification Using QGS and 4D-MSPECT

The reoriented datasets were loaded into fully automated QGS (4,5), which first determined the maximal-count midmyocardial surface. Next, rays were subtended normally to it and count profiles were extracted for each ray. Endocardial and epicardial boundaries were then estimated using an asymmetric gaussian fit of the count profiles.

Unlike QGS, the semiautomated 4D-MSPECT software (6) processes data on the basis of a 2-dimensional gradient image from which the initial estimates of the ventricle are made. Thereafter, a series of 1- and 2-dimensional weighted splines are used to refine the endocardial and epicardial surface estimates. In 10 patients, 4D-MSPECT did not adequately detect the estimated basal plane, and the basal planes were therefore adjusted using the interactive quality assurance module.

QGS failed to accurately detect myocardial borders in 1 patient, and 4D-MSPECT, in 2; thus, the algorithms together provided the EDV, ESV, and LVEF values for 42 patients in common.

cMRI Methodology and Data Analysis

Within a median of 2 d, all patients underwent cMRI on a 1.5-T Gyroscan ACS-NT (Philips Medical Systems) using a balanced fast-field echo sequence also known as true fast imaging with steady-state precession (12). Integrated SENSE (SENSitivity Encoding) technology allowed superior image quality because of the shorter acquisition time and the higher contrast-to-noise ratio. Patients were imaged in end-expiration using a dedicated cardiac synergy coil and breath-hold technique, with repetition and echo times of 3.1 and 1.5 msec, respectively, a flip angle of 65°, and a matrix of 256 × 256 (field of view, 350–400 mm). Twenty phases were obtained per cardiac cycle. The 10- to 15-min scanning protocol included the vertical and horizontal long axes of the left ventricle, on which the planning of the short axis was then based. From base to apex, the whole left ventricle was acquired with slices of 8-mm thickness, resulting in full-volume datasets for all patients.

Left ventricular function was assessed by determining the EDV, ESV, and subsequent LVEF using the cardiac analysis software package provided by the manufacturer. The examiner did not know the results of the gated ¹⁸F-FDG PET analysis. After the cardiac base and apex were determined, the first gate in each series was defined as the end-diastolic phase, and the image with the smallest ventricular volume was defined as the end-systolic phase. The endocardial borders of both heart phases were traced manually, with the trabeculation and the papillary muscles segmented as part of the myocardium. Both EDV and ESV were then automatically computed by summing the cross-sectional areas contained by the endocardial borders of all analyzed short-axis slices (10,11). The LVEF, expressed as a percentage, was calculated as the stroke volume divided by the EDV.

Statistical Analysis

Statistical analysis was done using SPSS 10 (SPSS Inc.) and Origin 6.1 G (OriginLab Corp.) software. The data are shown as mean ± SD. The mean values of EDV, ESV, and LVEF were tested for significance using a t test for paired samples while applying Bonferroni correction for multiple comparisons. A normal distribution was shown for all parameters (Kolmogorov-Smirnov test). The degree of agreement was tested according to the method of Bland and Altman (17); the Bland-Altman limits (mean of the differences ± 2 SDs of the differences) are shown in the figures. Pearson correlation coefficients were also calculated. P < 0.05 was accepted as significant.