Researchers ID sources of error for CT-measured cardiac function
Despite the excellent spatial and contrast resolution of CT, the modality comes up a bit short when calculating left ventricular size and function compared with MRI. Researchers from the University of Chicago were able to explain why this happens and how to solve the problem.
Dianna M.E. Bardo, MD, and colleagues sought to evaluate three potential sources of error in the volumetric analysis of CT data: accurate identification of end systole, insufficient number of slices covering the left ventricle (LV), and excess noise—as a result of tube current modulation during data acquisition (J Cardiovasc CT 2008;4:222-230).
They obtained CT images in 28 patients (Brilliance 64, Philips Healthcare) and reconstructed them at 10, 20, 33, and 100 phases per cardiac cycle. For each number of phases, they measured end-systolic volume (ESV) and end-diastolic volume (EDV) by semi-automatically tracing LV endocardial boundaries in the short-axis slice. They then calculated the ejection fraction (EF).
The investigators found that the true timing of end systole can be misidentified because of the small number of phases, such as routinely reconstructed from CT data (typically 10). This error results in an overestimation of ESV and underestimation of EF in individual patients, according to the study.
The timing of end systole varied with increasing number of reconstructed phases, resulting in a gradual decrease in ESV from 118 percent of the reference ESV with 10 phases to 100 percent with 33 phases.
The authors recommend reconstruction at time intervals smaller than 5 percent of the RR-interval (>20 phases/cardiac cycles).
To save time, researchers suggest estimating the timing of end systole from the initial reconstruction and then gradually refining it by selectively reconstructing a small number of additional phases around this estimate.
“The additional time needed for reconstruction and analysis would be reasonable (<10 minutes) and justifiable when accurate assessment of LV size and function is of particular importance,” they wrote.
In addition, to elucidate the effects of the number of slices on LV volumes, Bardo and colleagues varied their number from 20 to 10, while keeping slice thickness fixed at 5 mm. They found that the number of slices used for volume calculation affected both ESV and EDV.
The investigators also found that a reduction in the number of short-axis slices caused a significant increase in EDV and ESV (4.2 and 6.4, respectively). Accordingly, they suggest tracing endocardial borders in more than 10 slices.
In an accompanying commentary, Sandra S. Halliburton, PhD, from The Cleveland Clinic Foundation, pointed out the importance of the study’s reasonable inter-observer and intra-observer variability values.
Bardo et al. used an ECG-gated tube current modulation protocol, which reduces radiation exposure (average, <5 mSv).
Such a strategy also results in noisier images and the effect of the increase in noise “on the accurate delineation of endocardial borders and the calculation of ventricular volumes is not known,” Halliburton said.
Haliburton concluded that the inter-observer and intra-observer variability values were low enough to suggest that volumetric measurements can reliably be made from tube current modulated data.
She suggested further that being able to conduct LV functional exams with a low-dose protocol makes the “knowledge gained from the present study about accurate and efficient measurement of LV volume all the more valuable.”
Co-author Victor Mor-Avi, PhD, told Cardiovascular Business News that no patient will ever be initially referred to cardiac CT to evaluate LV function because of the iodine contrast and radiation, “no matter how accurate the technique is.”
He added that these data are already in the coronary CTA images and “it is just a matter of taking the time to analyze them.”
Improved algorithms for automated endocardial boundary detection are being developed and tested for MRI, but the research for “CT is not as far along because its use for cardiac functional imaging is relatively new,” Mor-Avi said.
Dianna M.E. Bardo, MD, and colleagues sought to evaluate three potential sources of error in the volumetric analysis of CT data: accurate identification of end systole, insufficient number of slices covering the left ventricle (LV), and excess noise—as a result of tube current modulation during data acquisition (J Cardiovasc CT 2008;4:222-230).
They obtained CT images in 28 patients (Brilliance 64, Philips Healthcare) and reconstructed them at 10, 20, 33, and 100 phases per cardiac cycle. For each number of phases, they measured end-systolic volume (ESV) and end-diastolic volume (EDV) by semi-automatically tracing LV endocardial boundaries in the short-axis slice. They then calculated the ejection fraction (EF).
 | |
| End-systolic LV volume (ESV, top) and ejection fraction (EF, bottom), normalized by the corresponding reference ESV and EF obtained from 100 reconstructed phases. Data was averaged for 16 patients and plotted against the number of reconstructed phases (*p<0.05 versus 100 reconstructed phases). This figure shows how ESV and EF converge towards their reference values with a gradual increase in the number of phases. Source: Victor Mor-Avi, PhD, University of Chicago | |
 | |
| Example of short-axis slice of the left ventricle from base to apex shown with the endocardial borders, which were semi-automatically traced for LV volume calculation. Source: Victor Mor-Avi, PhD, University of Chicago | |
The timing of end systole varied with increasing number of reconstructed phases, resulting in a gradual decrease in ESV from 118 percent of the reference ESV with 10 phases to 100 percent with 33 phases.
The authors recommend reconstruction at time intervals smaller than 5 percent of the RR-interval (>20 phases/cardiac cycles).
To save time, researchers suggest estimating the timing of end systole from the initial reconstruction and then gradually refining it by selectively reconstructing a small number of additional phases around this estimate.
“The additional time needed for reconstruction and analysis would be reasonable (<10 minutes) and justifiable when accurate assessment of LV size and function is of particular importance,” they wrote.
In addition, to elucidate the effects of the number of slices on LV volumes, Bardo and colleagues varied their number from 20 to 10, while keeping slice thickness fixed at 5 mm. They found that the number of slices used for volume calculation affected both ESV and EDV.
The investigators also found that a reduction in the number of short-axis slices caused a significant increase in EDV and ESV (4.2 and 6.4, respectively). Accordingly, they suggest tracing endocardial borders in more than 10 slices.
In an accompanying commentary, Sandra S. Halliburton, PhD, from The Cleveland Clinic Foundation, pointed out the importance of the study’s reasonable inter-observer and intra-observer variability values.
Bardo et al. used an ECG-gated tube current modulation protocol, which reduces radiation exposure (average, <5 mSv).
Such a strategy also results in noisier images and the effect of the increase in noise “on the accurate delineation of endocardial borders and the calculation of ventricular volumes is not known,” Halliburton said.
Haliburton concluded that the inter-observer and intra-observer variability values were low enough to suggest that volumetric measurements can reliably be made from tube current modulated data.
She suggested further that being able to conduct LV functional exams with a low-dose protocol makes the “knowledge gained from the present study about accurate and efficient measurement of LV volume all the more valuable.”
Co-author Victor Mor-Avi, PhD, told Cardiovascular Business News that no patient will ever be initially referred to cardiac CT to evaluate LV function because of the iodine contrast and radiation, “no matter how accurate the technique is.”
He added that these data are already in the coronary CTA images and “it is just a matter of taking the time to analyze them.”
Improved algorithms for automated endocardial boundary detection are being developed and tested for MRI, but the research for “CT is not as far along because its use for cardiac functional imaging is relatively new,” Mor-Avi said.