| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | RSS | TABLE OF CONTENTS |
|
|
|
||||||||
|
||||||||||
Clinical Investigations |
Unité 494, Imagerie Médicale Quantitative, Institut National de la Santé et de la Recherche Médicale, Paris; Service Hospitalier Frédéric Joliot, Département de Recherche Médicale, Direction des Sciences du Vivant, Commissariat à l’Energie Atomique, Orsay; Services de Cardiologie et de Médecine Nucléaire, Centre Hospitalo-Universitaire Henri Mondor, Université Paris XII, Créteil, France; and Department of Medical Physics and Bioengineering, University College London, London, England
 |
ABSTRACT |
|---|
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Key Words: coronary flow reserve • factor analysis • regularization • H215O PET scans • coronary artery disease • idiopathic dilated cardiomyopathy
|
INTRODUCTION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Processing of H215O PET dynamic image sequences with factor analysis methods (10,11) was proposed to estimate the myocardial factor images directly from dynamic studies (12). The limitation of the use of this conventional factor analysis algorithm is the effect of noise in the raw dynamic data on the factor images. To overcome this limitation, another method has been proposed that is close to factor analysis and based on the linear reduction of signal in the sinogram space (13). The method requires the introduction of some a priori physiologic information that assumes that the kinetics of the right and left cavities can be derived from the lung kinetics. The myocardial factor image is then reconstructed from the appropriate myocardial sinogram by an iterative reconstruction method. Finally, MBF is estimated by applying the 1-compartment model to time–activity curves resulting from the ROIs that are manually delineated on myocardial factor images. This procedure has recently been validated by comparing, in a group of volunteers and patients, the results obtained by the 2 ways of estimating myocardial ROIs: myocardial factor images and a conventional method that subtracts C15O images from H215O images (14).
This study evaluated an alternative method, based on the spatial regularization of factor analysis of medical image sequences (FAMIS), that was recently proposed (15) to improve the quality of factor images for studies with a low SNR. The method applies the conventional algorithm of factor analysis on filtered backprojection reconstructed dynamic series without requiring an a priori physiologic assumption. Regularization of the factor images is the last step of the algorithm. For patients, the new processing method to estimate the myocardial perfusion reserve (MPR) was compared with invasive measurement of coronary flow reserve (CFR) using an intracoronary Doppler technique. For all subjects (patients and volunteers), the MPR obtained from regularized factor images (RFIs) was compared with the MPR obtained from FDG acquisition.
|
MATERIALS AND METHODS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
|
Four of the 10 patients had had at least 1 episode of acute congestive heart failure related to idiopathic dilated cardiomyopathy. The coronary arteriogram did not show significant stenosis, and no other recognized etiology was evident. At least 1 wk before entering the study, all 4 patients were clinically stable and were taking diuretics and angiotensin-converting enzyme inhibitors.
Ten healthy volunteers having normal echocardiography findings were used as control subjects. They underwent the whole PET procedure but not the Doppler study. These volunteers were age-matched and sex-matched to the patients.
Study Protocol
The Doppler study was performed 1 d before the PET study. Neither clinical modification nor therapy change occurred between the 2 studies. All subjects refrained from oral intake of methyl xanthines, including caffeine, on the day of the studies so as not to diminish the effects of dipyridamole. The protocol was approved by the local ethical committee. Each subject gave informed consent.
Intracoronary Doppler Measurement of CFR.
After diagnostic catheterization, an additional dose of 5,000 U heparin was given intravenously, and an 8-French guiding catheter was positioned in the main left coronary artery. The electrocardiogram and mean arterial pressure were monitored continuously. A 3-French Doppler catheter (NuVel; NuMed, Inc., Hopkinton, NY) with a side-mounted crystal of 20 MHz was advanced through the guiding catheter into the proximal segment of the LAD remote to any large branch. The Doppler signal was transmitted to a velocimeter (MDV 20; Millar Instruments, Inc., Houston, TX) and recorded on an ES 1000 recorder (Gould Instrument, Inc., Ballainvilliers, France). The position of the catheter and the range of the sample volume were adjusted to obtain a high-quality signal, as assessed by both audio and graphic controls, and remained unchanged thereafter.
To ensure that the PET measurement of coronary reserve could be performed in similar conditions, no nitrate was injected through the guiding catheter before the Doppler assessment of vasodilator reserve. Dipyridamole (0.80 mg/kg) was infused intravenously at a rate of 10 mg/min. Coronary blood-flow velocities were continuously recorded during the 10 min after the end of dipyridamole infusion. The coronary reserve value was calculated as the mean velocity measured between the fifth and tenth minutes divided by the rest velocity.
After completion of coronary reserve measurement, coronary angiography was performed to evaluate the effect of dipyridamole on coronary artery diameter through a quantitative densitometric analysis of coronary diameter (TSI system; Troütement Synthése Image, Inc., Marne-la-Vallée, France).
PET Acquisition.
The subjects were positioned in a TTV03 time-of-flight PET scanner (LETI-CEA, Grenoble, France). Transmission scans were obtained with a retractable 68Ge ring source and used for subsequent attenuation correction of the emission scans. The electrocardiogram was monitored continuously during the examination and 30 min before and 60 min after the PET scan. Blood pressure was measured before injection and every 2 min during the whole examination. The positioning of the heart in the center of the field of view was checked by recording a 5-min transmission scan before the examination. Correct positioning was maintained throughout the study by using laser beams and by marking the skin of the subject’s torso.
MBF experiments consisted of intravenous bolus injections of H215O (11.1 MBq/kg). The experimental protocol included 2 injections with a 20-min delay to allow for 15O decay: one at baseline conditions and another 4–6 min after intravenous injection of dipyridamole (0.80 mg/kg, at a rate of 10 mg/min). Data were acquired in list mode during the 5 min after arrival of the blood radioactivity in the left ventricular cavity.
The subjects finally underwent an FDG PET study. They were given an oral 100-g glucose load 1 h before FDG imaging. After intravenous injection of an FDG concentration of 3.7 MBq/kg, data were acquired in list mode during 60 min.
PET Data Analysis
Image Reconstruction.
The H215O dynamic perfusion studies of 39 scans (15 x 4 s and 24 x 10 s) were reconstructed using a backprojection algorithm with a Hanning filter (cutoff frequency, 0.5 mm-1). Images were corrected for attenuation, random events, dead-time losses, and scattered radiation (16). A 20-min static FDG image recorded 40 min after injection was reconstructed using the same procedure. From 2 to 4 consecutive slices were considered for each scan.
Estimation of RFIs.
FAMIS was applied to dynamic reconstructed images, according to previously described algorithms (17,18), to process the whole set of slices simultaneously. Three factors (f1(t), f2(t), and f3(t), with t being the time variable) and 3 factor images (a1(p), a2(p), and a3(p), with p being the pixels of the images) were estimated from the image sequence. They were associated, respectively, with the right cavities, the lungs and left cavities, and the tissues, including myocardium and liver (Fig. 1).
|
 | (Eq. 1) |
The contribution ck of the kth factor image was defined as the part of the total information that the kth factor image represented in the image sequence such that:
 | (Eq. 2) |
Spatial regularization of factor images was achieved using a previously described algorithm (15). The method consists in minimizing a criterion E, being the sum of 2 terms. The first term, E1, represents the adequacy to the raw data, which is minimized by FAMIS; the second term, E2, introduces some a priori information about the factor images, assuming a regional smoothness of these images. The term E2 is mathematically defined as follows:
 | (Eq. 3) |
being a convex function to account for intensity variation between each pair of neighboring pixels. To avoid a blurring effect at the frontier between adjacent regions, quadratic function is replaced with the following function, h:
 |
The function h is quadratic for small variations in intensity (less than ) and linear for large variations (15).
The coefficient µ is set to adjust the trade-off between the fit to the raw data and the a priori information. The 3 terms in Equation 3 corresponding to the 3 factor images are weighted by the inverse of the contribution of each factor image to take into account the lower SNR of the factor images having the smaller contributions.
First trials on simulations showed that the regularization procedure improved factor images from both qualitative and quantitative points of view (15). Moreover, there was still some flexibility in the choice of the regularization parameters. The method was applied to real H215O myocardial studies by determining regularization coefficients from the simulation of a dynamic myocardial PET study (15). A value of 0.26 was chosen for the coefficient µ, and the parameter was set at 1% of the maximum value in the factor images. RFIs were computed slice by slice for each patient on both baseline and stress H215O studies (Fig. 2).
|
|
Two sets of MPR estimates were obtained: FAMIS-Op1 MPR, which was obtained from the ROIs that were drawn on the RFIs, and FDG MPR, which was obtained from the ROIs that were drawn on the FDG scans. A second estimation of MPR (FAMIS-Op2 MPR) was performed by a second operator for patient studies only.
Statistical Analysis
Parameters are expressed as mean ± SD. They were compared using paired or unpaired Student t tests when appropriate. A correlation coefficient, assuming a linear regression, was calculated for paired variables. The level of statistical significance was set at 0.05. Bland–Altman representation (22) was used to account for the absence of any absolute reference method.
|
RESULTS |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
No significant change in coronary artery diameter occurred after dipyridamole infusion. Individual measurements of intracoronary Doppler CFR are listed in Table 1. The mean CFR was 1.60 ± 0.33 for patients with idiopathic dilated cardiomyopathy and 1.82 ± 0.70 for patients with LAD stenosis.
CFIs and RFIs were computed systematically. The effect of regularization can be seen by comparing Figures 1 and 2. Table 2 shows the SNRs, which were estimated as the mean value of intensity divided by the SD of intensity, inside each myocardial region of CFIs and of RFIs. In all cases, regularization improved the SNR. The mean relative improvement in SNR (defined as the difference between the SNR of CFIs and the SNR of RFIs divided by the SNR of CFIs) was 70% for the global patient population, 81% for the group with idiopathic dilated cardiomyopathy, and 63% for the LAD group. Table 2 shows that the quality of myocardial factor images was generally worse for stress studies; the mean SNR inside myocardial regions of CFIs was 2.59 ± 0.49 for baseline studies and 2.29 ± 0.58 for stress studies. A paired t test proved that the difference was significant (P < 0.005). Regularization of stress images significantly improved SNR, which was 3.69 ± 1.10 (P < 0.00001). However, this SNR remained significantly less than the SNR in baseline RFIs, which was 4.59 ± 1.11 (P < 0.002).
|
For delineation of the left ventricular and myocardial ROIs from FDG scans, the static 1-h images were used without reference to the factor images. Individual MPR values obtained from the anteroseptal region defined on the FDG scans are shown in Table 1. The mean values (±SD) of MPR were 1.77 ± 0.36 for patients with idiopathic dilated cardiomyopathy, 1.87 ± 0.59 for patients with LAD stenoses, and 3.61 ± 0.68 for healthy volunteers.
Comparison Between CFR Measurements and Different MPR Measurements
For the 10 patients, 4 sets of reserve values were available (Table 1). Friedman ANOVA by ranks showed no significant difference among the methods of measurement. Results obtained using the RFIs correlated significantly with results obtained using the intracoronary Doppler method (Fig. 4) (y = 1.17x - 0.30; r = 0.97 for the first operator; y = 1.35x - 0.54; r = 0.95 for the second operator). The Bland–Altman representation of experimental data (Fig. 5) shows that the discrepancy between RFIs and intracoronary Doppler methods (0.37) was reduced when compared with the discrepancy between FDG and intracoronary Doppler methods (0.57). After removal of the outlier, the discrepancy between FDG and Doppler dropped to 0.42.
|
|
For the 10 healthy volunteers, 2 sets of MPRs were available. No significant difference was found between them. For the global population, a highly significant correlation (y = 0.99x; r = 0.93) was found between the MPR obtained from RFIs and the MPR obtained from FDG scans (Fig. 6).
|
|
|
DISCUSSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
Hermansen et al. (13) recently proposed another approach to generate high-quality myocardial factor images. Our method differs in several points. First, we analyzed the reconstructed images, not the sinograms. Second, no a priori physiologic information was required to estimate factors and factor images; in particular, we did not need a model to estimate heart cavity kinetics from lung kinetics. Third, 3 complementary factor images (right cavities, left cavities, and tissue) were generated, and all were useful for delineating myocardial ROIs; indeed, their superimposition avoided the potentially large contamination of myocardial regions by the cavities. Finally, standard and time-efficient algorithms for reconstruction (filtered backprojection) and factor analysis could be used. The main originality of the algorithm presented here was the final regularization of the factor images.
The clinical validation we present supplements clinical validations previously reported (13,14), which used the method of Hermansen et al. (13). Indeed, baseline MBF studies of ischemic patients (13) and MPR measurements of volunteers (14) have been reported, but not MPR measurements of patients. Moreover, the validation was compared with an independent method, the intracoronary Doppler technique.
For patients with idiopathic dilated cardiopathy, the MPR values (1.55 ± 0.36; range, 1.13–1.97) were close to those published in the literature; a mean value of 1.7 ± 0.21 was recently reported (23). For patients with coronary artery stenoses (range, 71%–91%), the mean MPR value was 1.87 ± 0.59 (range, 1.09–3.34). Similar values for patients with coronary artery stenoses have been reported (24,25). The low MPR values in the reference normal territory (non-LAD region) of CAD patients were already reported (25). A possible explanation is that an abnormal flow reserve in angiographically normal territories could represent early functional abnormalities of vascular reactivity (25).
Factor analysis was successfully applied to myocardial perfusion PET studies to estimate input function from small hearts using FDG (26) and to estimate myocardial factor images with good precision using 13N ammonia (27). H215O studies are more difficult to process because of the low SNR and the free diffusion of the water. Our study shows that regularized FAMIS applied to MBF H215O studies enabled the generation of high-quality factor images on which myocardium could be delineated accurately and separated from heart cavities and lungs. During stress studies, the volume of displacement of the myocardium is increased. Because the acquisitions are not gated by the electrocardiogram and the myocardial contraction is larger, the same quantity of tracer is distributed in an apparently larger volume. For this reason, the SNR of dipyridamole factor images is lower than the SNR of baseline factor images. However, regularization, which largely improves the SNR of both baseline and dipyridamole factor images, always makes possible myocardial segmentation.
Although RFIs are useful for the segmentation of the myocardial wall, factors cannot be directly used for the modeling step. Estimated factors are not equivalent to myocardial time–activity curves (Fig. 1), for several reasons. The initial undersampling of pixels in the image sequence (6 x 6 clusters in myocardial studies) generates mixed kinetics. The aggregation improves SNR in factor estimation, but as a result, the factors are contaminated by each other. Moreover, because only 3 factors are estimated, the lung kinetics are not extracted, and these factors are also contaminated by the lung kinetics. As a consequence, factors and factor images cannot be used directly to extract quantitative information. The myocardial factor image is not strictly equivalent to a tissue blood-flow image, even if the myocardial factor image is related to this blood-flow information.
Uncontaminated kinetics could be estimated by FAMIS using adequate physiologic constraints. Such an estimation would require a complete modeling of the kinetics of the study, including the circulation in the right and left cavities and in the lungs, which is beyond the scope of this article. However, this step is a prerequisite for obtaining RFIs that are of true diagnostic image quality. In this study, the value of the RFIs was assessed only in the crucial step of performing an accurate myocardial segmentation for H215O studies.
The aim of this study was to estimate MBF or coronary reserve using H215O dynamic PET scans alone. The advantages of such a modified procedure are obvious and significant: a reduction in the radiation dose to the patient, a shortened acquisition time, and a consequent reduced examination cost. Furthermore, the effects of possible patient motion between the dynamic studies and the supplementary scan are removed. This possible motion results in an error in the positioning of ROIs in the dynamic studies and can explain the larger differences that were observed between FDG and intracoronary Doppler studies.
Different abnormalities have been studied, and the method has proven robust. However, patients with 3-vessel disease that includes large, hypoperfused necrotic territories should be specifically studied and may reveal an intrinsic limitation of the method caused by poor count statistics in H215O PET scans.
Despite the suggestion that MRI may be used to estimate a coronary reserve index (28), the potential field of applications for H215O PET scans remains wide because PET ensures that absolute values of regional peak and baseline MBF will be determined. Absolute myocardium flow measurements ensure a direct evaluation of MBF reserve that reflects the ability of vasculature to increase flow maximally in response to a hyperemic stimulus. The 2 techniques (PET and Doppler) are not strictly equivalent, because PET measurements are directly related to the myocardium, whereas Doppler measurements are performed on coronary arteries. The absolute quantification of MBF or of MPR can be used in pharmacology trials to measure the impact of drugs. The opportunity to measure MPR noninvasively is an advantage in following up the effect of medical therapy and of major therapeutic interventions, such as revascularization procedures, in ischemic patients. Moreover, with PET, one can simultaneously assess perfusion and metabolism or perfusion and receptor kinetics.
|
CONCLUSION |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
|
ACKNOWLEDGMENTS |
|---|
|
FOOTNOTES |
|---|
For correspondence or reprints contact: Frédérique Frouin, PhD, U494 INSERM, CHU Pitié-Salpêtrière, 91 Blvd. de l’Hôpital, 75634 Paris Cedex 13, France.
|
REFERENCES |
|---|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONCONCLUSIONREFERENCES |
|---|
This article has been cited by other articles:
|
|
 |
|
  A. Cassar, P. Chareonthaitawee, C. S. Rihal, A. Prasad, R. J. Lennon, L. O. Lerman, and A. Lerman Lack of Correlation Between Noninvasive Stress Tests and Invasive Coronary Vasomotor Dysfunction in Patients With Nonobstructive Coronary Artery Disease Circ Cardiovasc Interv, June 1, 2009; 2(3): 237 - 244. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 T. O. Kiviniemi, A. Snapir, M. Saraste, J. O. Toikka, O. T. Raitakari, M. Ahotupa, J. J. Hartiala, M. Scheinin, and J. W. Koskenvuo Determinants of coronary flow velocity reserve in healthy young men Am J Physiol Heart Circ Physiol, August 1, 2006; 291(2): H564 - H569. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. S. Lee, D. S. Lee, J. Y. Ahn, J. S. Yeo, G. J. Cheon, S.-K. Kim, K. S. Park, J.-K. Chung, and M. C. Lee Generation of Parametric Image of Regional Myocardial Blood Flow Using H215O Dynamic PET and a Linear Least-Squares Method J. Nucl. Med., October 1, 2005; 46(10): 1687 - 1695. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. A. Lammertsma Myocardial Perfusion in 3 Dimensions J. Nucl. Med., August 1, 2002; 43(8): 1041 - 1043. [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | RSS | TABLE OF CONTENTS |
| JOURNAL OF NUCLEAR MEDICINE TECHNOLOGY | THE JOURNAL OF NUCLEAR MEDICINE |
