TO THE EDITOR:
We write to point out several issues related to the article by Vallejo et al. (1) on experimental validation of gated SPECT for determining left ventricular volumes and ejection fraction.
Although one of the stated goals of this article was validating left ventricular ejection fraction (LVEF) and volume determinations by quantitative gated SPECT (QGS) against an MRI standard, Vallejo et al. (1) did not quote six separately published articles and abstracts from independent groups showing excellent agreement in humans between QGS and MRI LVEF (r = 0.85–0.94), end-diastolic volume (r = 0.81–0.95), and end-systolic volume (r = 0.90–0.97). Limitations on the number of references allowed in Letters to the Editor prevent us from explicitly quoting every one of these studies (2–4), but the reader can find the complete list at http://www.csmc.edu/aim/vallejo. By contrast, Vallejo et al. found poor correlation for LVEF in one canine model.
Additionally, the contention of Vallejo et al. (1) that QGS overestimates volumes in the presence of perfusion defects contradicts 12 published findings by 8 independent investigators studying humans. Again, we explicitly quote six of those studies (5–10), and the complete list can be found at http://www.csmc.edu/aim/vallejo. Interestingly, these studies are not listed in the bibliography of the article.
It is possible that the discrepancies between the findings of Vallejo et al. (1) and those in published QGS articles are caused by the fact that they imaged dogs. Animal studies are, of course, a valuable tool when patient data are unavailable, but they often may not be relevant to clinical practice because of the obvious differences in acquisition, positioning, type and mode of (simulated) occlusion, and degree of overlap from the liver. The last is likely to have been of particular importance in the animal model and may explain many of the problems noted with quantitation in this article. Unfortunately, the omission by Vallejo et al. of the failure rate of QGS in their canine population (both in the totally automatic mode and in the manual mode) makes it difficult to assess the specific problems they may have encountered. Also, no images are shown of studies in which successful and unsuccessful segmentation was attained by QGS.
Finally, it might have been appropriate to disclose that at least one of the authors and one of the consultants acknowledged in the article by Vallejo et al. (1) are involved in the development of a commercially available software package for the quantification of ejection fraction, wall motion, and wall thickening from gated SPECT images. The software, distributed under the name Wackers–Liu CQ (Eclipse Systems, Branford, CT), has been licensed to several nuclear medicine companies. It competes directly with QGS, the Cedars–Sinai (Los Angeles, CA) software tested in the article.
REFERENCES
REPLY:
We appreciate Germano and Berman’s interest in our recent publication (1), which evaluated their commercially available software program (QGS) for evaluation of gated SPECT in experimental models. They are correct in stating that one of our three aims was to compare the accuracy of QGS for estimation of left ventricular end-diastolic (ED) and end-systolic (ES) volume in our experimental model with that derived using a true three-dimensional analysis of MR images. We also explored the effect of a perfusion defect on this relationship. Admittedly, we put their algorithm to a real test by analyzing a group of hearts with large, dense anteroapical perfusion defects. However, this type of anteroapical defect is often seen in patients after acute anterior myocardial infarction. Obviously, all experimental animal models have strengths and weaknesses. One potential weakness of our study was that the position of the dog heart in relation to the liver was slightly different from that in humans. The healthy dog heart may also be slightly smaller than the average human heart. This potential limitation was stated in our article. However, coronary occlusion in the dog generally causes immediate left ventricular dilatation. The average QGS ES volume we observed in our dogs (group I, without defects: 24 ± 3 mL; group II, with defect: 41 ± 5 mL) was similar to the ES volumes (males: 37.4 ± 13.7 mL; females: 21 ± 11 mL) reported by Kang et al. (2) in patients using QGS. This study was referenced in our article.
Germano and Berman brought to the reader’s attention six additional references (three abstracts, three manscripts) comparing QGS volumes with MR-derived volumes. All but two of the references were published after submission of our article in February 1999. The two references published before submission of our article (3,4) represent the same study; the former is the abstact and the latter is the article. This study did not evaluate true three-dimensional volume from MR images but simply applied Simpson’s rule to short- and long-axis dimensions derived from MR image sets. This situation represents an important limitation of that study for estimation of true left ventricular volume using MRI. A recent study (5) comparing two-dimensional and three-dimensional echocardiography and MRI suggested that performing true three-dimensional analysis of left ventricular volume is more important than the method by which the data are derived. Only one of the cited studies (6) computed true three-dimensional volume using MRI. Although each of the clinical studies cited by Germano and Berman showed a reasonable correlation between QGS and MR-derived volumes, all of the studies revealed a large SEE, ranging from 7 to 29 mL. This result represents a significant error, considering the average left ventricular volumes.
Germano and Berman also cited 11 additional clinical studies supporting the accuracy of QGS for estimation of left ventricular volume in the presence of a perfusion defect. Again, 6 of the 11 cited studies were published only in abstract form. All of the studies involved relatively small numbers of patients (population sizes, 20–72 patients; average, 43 patients). These studies compared QGS with a host of other methods, including echocardiography, MRI, equilibrium radionuclide blood-pool imaging, and radionuclide first-pass imaging. Most of these studies focused on the comparison of ejection fraction and not volumes. These clinical studies showed inconsistent differences in ED and ES volumes. None of these referenced studies specifically addressed the issue of perfusion defect size in relation to calculation of ventricular volumes.
Our group recently evaluated QGS in 400 unselected patients for estimation of ejection fraction in comparison with radionuclide first-pass imaging (7). The correlation of left ventricular ejection fraction (LVEF) derived using QGS and first-pass imaging was only fair (r = 0.66, SEE = 12%). The automated program failed in 9% of the studies. When the fully automated program worked, the correlation was better (r = 0.74); however, the SEE remained high (SEE = 10%). In this large clinical study, we observed a better correlation for high-count images (r = 0.81, SEE = 9%) than for low-count images (r = 0.61, SEE = 11%). This observation was similar to our experimental canine study. As expected, we observed a better correlation of QGS and first-pass radionuclide angiography in very large hearts, with ED volumes > 104 mL. We also evaluated the effect of quantitative perfusion defect size on estimation of ejection fraction. Unlike the experimental study, in our clinical study we observed a better correlation of first-pass and QGS LVEF in patients with the largest defects. However, these patients also had the largest left ventricular sizes, making it difficult to separate the potentially opposing effects of myocardial perfusion defect size and left ventricular size. In any event, we would not necessarily expect to see the same differences between QGS and MRI and other imaging modalities. In fact, the studies referenced by Germano and Berman using MRI showed conflicting results regarding over- and underestimation of volumes.
As stated on page 877, paragraph 4, of our article (1), the QGS program failed in 21% of our canine images, which was more frequently than observed in our clinical study. This failure was generally associated with adjacent background activity. Our original submission included a figure of canine images obtained 15 and 45 min after injection of a radiotracer. We were asked to remove the figure from our article by the editor to reduce the number of figures. This figure showed the excellent image quality of our canine studies and visually acceptable QGS-defined endocardial and epicardial contours, using the fully automated program. In this example, serial images obtained only 30 min apart yielded significantly different LVEFs (15 min: LVEF = 41%; 45 min: LVEF = 52%). We found this particular observation quite troubling.
When our article was accepted for publication in August 1999, we did not feel that a potential conflict of interest existed for any of the authors. In August 1999, our group did enter into an agreement with a vendor (Eclipse Systems, Branford, CT) to commercialize our software (Wackers–Liu CQ) for quantification of static SPECT myocardial perfusion images. We receive royalties for our program for quantitative analysis of static SPECT images. However, when we entered into this agreement, our program did not include an algorithm for calculation of LVEF or left ventricular volume. In January 2000, well after acceptance of our revised article, we began development of a program for the calculation of left ventricular volume and LVEF. These programs remain under development and testing and currently await Food and Drug Administration approval. We plan to put our own program, as well as the other available programs for analysis of gated SPECT, through the same rigorous testing.