Measurement of Ventricular Volumes and Function: A Comparison of Gated PET and Cardiovascular Magnetic Resonance

CMR Methodology

Nine patients were imaged with a Picker Edge 1.5-T scanner (Picker, Cleveland, OH), using the body coil, ECG triggering, and parameters previously described (3). Briefly, the cardiac short axis (SA) was determined from 3 scout images of the LV: the transverse, vertical long axis, and breath-hold diastolic horizontal long axis. The basal SA slice was positioned just forward of the atrioventricular ring, and all subsequent breath-hold cines were acquired in 1-cm steps toward the apex. A breath-hold segmented gradient-echo fast low-angle shot sequence was used for each of the contiguous SA slices encompassing the LV. Image analysis was performed on a personal computer using in-house developed software (CMRtools, Imperial College, London, U.K.). The reproducibility of this technique in our center has been previously published: The interstudy percentage variability was 2.5% for end-diastolic volume (EDV), 3.1% for end-systolic volume (ESV), and 4.8% for ejection fraction (EF) (3).

PET Methodology

PET studies were performed on the ECAT EXACT 3-dimensional positron tomograph (CTI, Knoxville, TN). A transmission scan was obtained using a 150-MBq ¹³⁷Cs point source. The emission data were acquired using inhaled C¹⁵O (half-life, 2.06 min). Labeling of red blood cells occurs through the formation of carboxyhemoglobin, which remains limited to the intravascular space. The gas (1.5 MBq/mL) was delivered at a rate of 500 mL/min through a face mask for 4 min (effective dose equivalent, 1.15 mSv). Data acquisition commenced immediately and continued for 11 min. This scanner allows data acquisition in list mode synchronized with the ECG R wave, and thus retrospective (or after acquisition) off-line rebinning of data into datasets corresponding to time frames per R−R interval (gates). Eight gates per cardiac cycle were used. The data were corrected for attenuation and scatter and were reconstructed by dedicated array processors and a reprojection reconstruction algorithm using a Hann filter (cutoff at the Nyquist frequency).

A threshold-based edge detection algorithm was used to calculate LV and RV volumes (4). The gated images were averaged together to provide a static blood-pool image. This static image was used to define SA reslice parameters. Both the static and gated images were then resliced to the SA view. The static image was used to identify those planes contributing to the RV and LV; any planes inferior or superior to the ventricles were discarded from the gated image. The atrioventricular valve plane was identified as the plane where the gap between the LV and RV disappeared. LV and RV analysis was performed individually. The maximum value for the image was determined by applying 4 circular regions of interest to 4 image planes containing the ventricle being studied. The voxels within the 4 regions were treated as a volume of interest (VOI). The maximum voxel value of the VOI divided by 2 (half maximum value) was used as the threshold, which defined the cutoff value of the ventricular cavity boundary. Signal contribution of the myocardial vasculature was assumed to be 10% of the blood pool, and the thresholds were scaled accordingly: threshold = half maximum + (half maximum × background fraction). The threshold values obtained were used with an interactive edge detection program to automatically determine the ventricular chamber boundary for each gate. This area multiplied by the plane thickness results in a plane volume. EDV and ESV were calculated by summing all the plane volumes at end-diastole and end-systole, respectively. Stroke volume (SV) = EDV - ESV. EF = SV/EDV. Interobserver variability is 1.1% for EDV and 1.8% for ESV using this technique (4).