Quantification and data optimisation of heart and brain studies in conventional nuclear medicine Dobbeleir, André Alfons

University of Groningen Quantification and data optimisation of heart and brain studies in conventional nuclear medicine Dobbeleir, André Alfons IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below. Document Version Publisher's PDF, also known as Version of record Publication date: 2006 Link to publication in University of Groningen/UMCG research database Citation for published version (APA): Dobbeleir, A. A. (2006). Quantification and data optimisation of heart and brain studies in conventional nuclear medicine s.n. Copyright Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons). Take-down policy If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim. Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum. Download date: 16-10-2018

13 2. Determination of left ventricular ejection fraction by first pass and gated SPECT studies. 2.1.Performance of a single crystal digital gamma camera for first pass cardiac studies. A. Dobbeleir 1, P.R. Franken 1, H.R. Ham 2, C. Brihaye 3, M. Guillaume 3, F.F. Knapp 4 and J. Vandevivere 1 1 Division of Nuclear Medicine, Middelheim General Hospital Antwerpen, 2020 Belgium, 2 Division of Nuclear Medicine, St Peter s Hospital, 1000 Bruxelles, Belgium, 3 Cyclotron Research Center, University of Liège, Belgium, 4 Nuclear Medicine Group, Health and Safety Research Division, Oak Ridge National Laboratory (ORNL), Oak Ridge, TN 37831-0622, USA Nuclear Medicine Communications 1991; 12: 27-34. Summary First pass radionuclide angiocardiography (FPRNA) has gained increasing interest because of the development of new 99 Tc m -labelled perfusion agents and of new 191 Os/ 191 Ir m generator systems. The aim of the study was to evaluate the performance capacities of a small field of view crystal digital gamma camera for 99 Tc m and 191 Ir m at high count rates. The camera dead time for 99 Tc m (window 30%) was well corrected up to 300 kcps in fast acquisition mode using the relative decrease of a small shielded reference source. Using the decaying activity method for 191 Ir m the non-linearity response of the gamma camera was corrected by an 191 Os reference source up to 210 kcps at 70 kev, 75 kcps at 129 kev and 320 kcps including both peaks. Saturation count rates were respectively 270 kcps, 150 kcps and 420 kcps and high count rate resolution (FWHM) 9.0, 7.3 and 10.3 mm. Since the accuracy of the first pass measurements is more sensitive to count rate than to spatial resolution the 50-150 kev window was chosen for clinical studies. In data obtained from 32 ECG gated FPRNA patient studies, the whole field of view count rate during the left ventricular phase ranged from 100 to 250 kcps with 80 to 120 mci (2960-4400 MBq) of 191 Ir m and 100 to 180 kcps with 20 to 25 mci (750-925 MBq) of 99 Tc m red blood cells permitting for both tracers accurate non-linearity correction. Introduction First pass radionuclide angiocardiography (FPRNA) has recently gained increasing interest for measuring left ventricular function. Firstly, because of the availability of new 99 Tc m -labelled myocardial perfusion agents allowing simultaneous assessment of myocardial perfusion and function [1, 2]. Secondly, because of the development of new high performance 191 Os/ 191 Ir m generator systems [3-5] offering the opportunity to conduct rapid, repeat, multiple first pass studies of the cardiovascular system with the ultrashort half-lived 191 Ir m [6-8]. The aim of this study was to evaluate the performance capacities and the limitations of a single crystal digital gamma camera (SCDGC) with respect to the high count rates needed for accurate measurements of ventricular function with the FPRNA method.

14 Materials and methods 191 Ir m 191 Ir m is the daughter of 191 Os ( 191 Os:β - emission ; T 1/2 = 15,4 days) and decays with a half-life of 4,96 s to stable iridium emitting a gamma-ray at 129 kev and three X-rays at 63 kev, 16%; 65 kev, 28%; and 74 kev, 12%. The X-rays cannot be resolved with the Na crystal and appear thus as one single peak at about 69 kev (Fig 1.). In this study, 191 Ir m was produced by elution of a carbon-based 191 Os/ 191 Ir m generator system with ph 2, 0,9% NaCl solution containing potassium iodide and subsequently neutralized with a TRIS buffer. Details concerning the preparation and use of this generator system have been published elsewhere [3, 8, 9]. Data acquisition Data were acquired in the normal acquisition mode or in the fast acquisition mode with a small field of view (20 cm) SCDGC (APEX 215M, Elscint) equipped with a very high sensitivity, lowenergy, parallel hole collimator. In the fast mode, a higher number of counts can be acquired using a different electronic circuit integrating only the first 400 ns of the scintilation. Fig. 1. Spectrum of 191 Ir m measured with a gamma camera. Camera resolution The resolution of the camera for different energies was tested with 99 Tc m, 201 Tl and 191 Os point sources in the normal and in the fast acquisition modes [10]. 191 Os decays to 191 Ir m by β - emission without emitting photons. The FWHM, FW20M and FW10M were calculated in a 30% window centered over the 140 kev 99 Tc m photopeak, in a 40% window centered over the 70 kev 201 Tl photopeak and in the 50-100, 100-150 and 50-150 kev windows of the 191 Ir m spectrum. Count-rate linearity The linearity of the gamma camera for 99 Tc m was measured by placing an increasing number of small vials (1 cm diameter) on the collimator. The activities were measured using a dose calibrator

and no scattering material was used. Data were acquired in the fast mode with the energy of the pulse height analyser set at 140 kev with a 30% window. Increasing 99 Tc m activities were measured together with a 10 kcps activity shielded reference source of 99 Tc m placed in the field of view of the camera. The measured activity was corrected for the dead time of the camera by a software correction program based on the detected counts before the pulse height analyser (provided by the manufacturer) and by the relative decrease of activity of a reference source [11, 12]. The linearity of the gamma camera for 191 Ir m was tested using the decaying source method. For that purpose a bolus of 100 mci (3700 MBq) of 191 Ir m was extracted from the generator system using 15 ml of normal saline solution, collected in an extension tube and directly divided through a three-way stopcock into two 100 ml beakers placed on the collimator. The beakers contained a small quantity of water in order to obtain a distributed source. A 10 kcps activity shielded 191 Os source was placed on the camera as reference. Data were acquired in dynamic mode (25 frames per second) for 30 s. This procedure was repeated three times in the 50-100, 100-150 and 50-150 kev windows. Time-activity curves were then generated from regions of interest (ROI) drawn over the reference source, over the two beakers and over a region between the two beakers ROIs, the latter being used to estimate the relative amount of misplaced pile-up events [13, 14]. The activity curves of the two beakers were added together and corrected for camera dead time by the relative decrease of the reference source activity. The linearity response of the gamma camera was then established by comparing the dead time corrected activity curve to the theoretical decaying curve of 191 Ir m. For this purpose, a linear fit with a slope = -0.140 corresponding to the decay constant of 191 Ir m was applied on the low values of the corrected activity curve expressed in the natural log (Fig. 2). Accepting a 1% deviation as criteria, the limits of the linearity response of the system was determined for the above-mentioned windows of the 191 Ir m spectrum. Fig. 2. Linear fit with a slope of 0.140 is applied to the dead time corrected decay curve of 191 Ir m in order to determine the limits of linearity response of the system. 15 Patient studies First pass radionuclide angiocardiographic studies were obtained at rest in 32 patients with 80-120 mci (2960-4400 MBq) of 191 Ir m and a few minutes later with 20-25 mci (750-925 MBq) of 99 Tc m red blood cells. Pulse height analyser windows were set over the 50-150 kev windows for 191 Ir m

16 studies and over the 119-161 kev windows for the 99 Tc m studies. Data were collected in the fast mode in a 32 x 32 x 8 matrix (25 frames per second) for 30 s. Results Camera resolution The FWHM, FW20M and FW10M with 99 Tc m, 201 Tl and 191 Os point sources are given in table 1. Table 1. Spatial resolution on point sources of 99 Tc m, 201 Tl and 191 Ir m (50-100, 100-150 and 50-150 kev windows) in the normal (N) and fast (F) acquisition modes. 191 Ir m 99 Tc m 201 Tl 50-100 kev 100-150 kev 50-150 kev N F N F N F N F N F FWHM 6.0 7.3 6.5 8.6 6.9 9.0 6.4 7.3 7.7 10.3 FW20M 9.8 11.2 10.3 12.9 10.3 14.2 9.9 12.0 11.2 15.0 FW10M 12.4 14.2 13.3 16.3 12.5 16.3 12.0 14.2 13.3 18.1 As expected, the spatial resolution of the system was better for 99 Tc m than for 201 Tl in the normal as well as in the fast acquisition modes. In the latter, a small but consistent degradation of the resolution was observed with both isotopes. The resolution of the system for 191 Ir m depends obviously on the window selection. The resolution was similar to that of 201 Tl for the 50-100 kev window (FW20M 10.3 mm versus 10.3 mm) and similar to that of 99 Tc m for the 100-150 kev window (FW20M 9.9 mm versus 9.8 mm) while the largest window (50-150 kev) gave the largest FW20M (11.2 mm). Again the fast acquisition mode induced a degradation of the spatial resolution for all energy windows. This influence of window selection on the spatial resolution of the gamma camera was further observed in clinical studies comparing 191 Ir m FPRNA to 99 Tc m. The left ventricular ROI area averaged 164 ± 29 pixels in 99 Tc m studies and 192 ± 28 pixels in the 191 Ir m studies (P<0.0001). Camera linearity Using the relative decrease of the shielded point source activity as a reference, the camera dead time for 99 Tc m (with a 30% window) was corrected up to 80 kcps in the normal acquisition mode and up to 300 kcps in the fast acquisition mode, corresponding to true count rates of about 160 and 650 kcps, respectively. The software correction resulted in systematic undercorrection of the linearity response of the camera. The saturation count rate for 191 Ir m was 270 kcps in the 50-100 kev window, 150 kcps in the 100-150 kev window and 420 kcps in the 50-150 kev window. Using the decaying source method, the camera dead time was corrected with an error of less than 1% up to 210 kcps in the 50-100 kev window, 320 kcps in the 50-150 kev window, but only up to 75 kcps in the 100-150 kev window. The number of pile-up events was estimated from the ROI drawn between the two beakers. At bolus arrival up to 9% of the total measured activity in the camera field of view was related to misplaced events in the 100-150 kev window compared to 2.5% in the 50-100 kev window and 5.5% in the 50-150 kev window (Fig. 3). A 1% or less misplaced events were observed in the 50-100 kev window at the maximal count rate capacity of the camera (270 kcps), in the 50-150 kev window at 70% (300 kcps) of the maximal capacity, but in the 100-150 kev at only 40% (60 kcps) of the maximal capacity of the camera system. Patient studies The highest count rates in the WFOV and in the right and left ventricular ROIs during the first transit of 191 Ir m (50-150 kev window) observed in 3 of the 32 patients are given in table 2. In Patient 1, although left ventricular count rate was rather low the WFOV count rate during the right

phase of the transit was just below the maximal limit of accurate dead time correction of the system. In Patient 2, and certainly in Patient 3, the count rates during the right transit phase were out of the limits of dead time correction: the maximum WFOV count rate was reached at 0.3 and 2.6 s, respectively, after the passage of the bolus in the right ventricle, indicating saturation of the gamma camera and precluding simultaneous assessment of right and left ventricular function studies during a single injection of 191 Ir m. Table 2. Maximal count rates during 191 Ir m (50-150 kev window) FPRNA studies in three patients. FPRNA RV phase LV phase WFOV RV WFOV LV WFOV Patient (kcps) (kcps) (kcps) (kcps) (kcps) 1 306 201 306 51 132 0.0 2 400 278 398 99 212 0.3 3 441 187 380 195 331 2.6 Delay max FPRNAmax RV phase ( s) The maximal count rate in the 20 cm field of view of the camera observed in most patients during the left ventricular transit of the 191 Ir m bolus ranged between 100 and 250 kcps. The maximal count rate observed in those patients during the left ventricular transit of the 99 Tc m bolus ranged between 100 and 180 kcps. Left ventricular counts in the 40 ms end-diastolic image of the ECG-gated left ventricular representative cycle averaged 14.8 kcounts (range 5.3-30.3) with 191 Ir m and 10.2 kcounts (range 3.6-22.1) with 99 Tc m. Discussion The count rates observed during left ventricular FPRNA studies using 99 Tc m and 191 Ir m were within the limits of accurate dead time correction for this gamma camera system. Left ventricular counts were sufficiently high to measure left ventricular function accurately [15]. Window selection on the 191 Ir m spectrum with the pulse height analyser is of major importance for both camera resolution and linearity when performing studies with this tracer. Although the 100-150 kev window is associated with the best camera resolution, this selection is the worst with respect to count rate capacities and dead time correction because the relative low contribution of those photons to the total number of photons reaching the crystal and because of the pile-up events. Accurate dead time corrections with the reference activity source were obtained, in the 50-100 kev as well as in the 50-150 kev window, for count rates higher than those observed in patients during left ventricular first pass studies. Although the spatial resolution of the 50-100 kev was somewhat better than the 50-150 kev window, this latter was chosen for clinical studies because the accuracy of first pass measurements is known to be more sensitive to count rate than to spatial reslution [15]. Using this window, the number of counts in the left ventricular cavity with 191 Ir m were at least equal to those obtained with 99 Tc m in all patients. During the right ventricular transit phase of the bolus, the total count rate was obviously over the count rate capacities of the camera in most patients precluding simultaneous studies of the right en left ventricles during a single injection of 191 Ir m. In our patient population, the mean total activity in the 20 cm field of view of the camera was about 1.15 times higher during diastole than during systole, introducing a relative dead time correction of 1.03 to 1.04. On the other hand, for 191 Ir m, the relative decay correction between diastolic and systolic frames ranged between 1.04 and 1.06. As the linearity correction factor and the decay correction factor work in the opposite direction, an error of maximum 3% would be made on the ejection fraction not applying any correction. For large field of view gamma cameras one can expect a smaller relative dead time correction between diastolic and systolic frames due to a less varying total activity in the large field of view. 17

18 Fig. 3. Time-activity curves of a decaying 191 Ir m source (initial activity 100 mci) in the 50-100, 100-150 and 50-150 kev windows, respectively. For display purposes, the activity of the 191 Os reference source and of the misplaced events recorded simultaneously, were multiplied by a factor of 5. References 1. Sporn V, Perez Balino N, Holman BL et al. Simultaneous measurement of ventricular function and myocardial perfusion using the technetium-99m isonitriles. Clin Nucl Med 1988; 13: 77-81. 2. Baillet GY, Mena IG, Kuperus JH et al. Simultaneous technetium-99m MIBI angiography and myocardial perfusion imaging. J Nucl Med 1989;30: 38-44. 3. Brihaye C, Butler TA, Knapp FF Jr et al. A new osmium-191/iridium-191m radionuclide generator system using activated carbon. J Nucl Med 1986; 27: 380-7.

4. Packard AB, Treves ST, O Brien GM, Lim KS. An osmium-191/iridium-191m radionuclide generator using an oxalato osmate parent complex. J Nucl Med 1987; 28: 1571-6. 5. Issachar D, Abrashkins S, Weinigerr J et al. Osmium-191/iridium-191m generator based on silica gel imregnated with tridodecylmethylammonium chloride. J Nucl Med 1989; 30: 538-41. 6. Heller GV, Treves ST, Parker JA et al. Comparison of ultrashort-lived iridium-191m with technetium-99m for first pass radionuclide angiocardiographic evaluation of right and left ventricular function in adults. J Am Coll Cardiol 1986; 7: 1295-302. 7. Hellman C, Zafrir N, Shimoni A et al. Evaluation of ventricular function with first pass iridium-191m radionuclide angiography. J Nucl Med 1989; 30: 450-7. 8. Franken PR, Dobbeleir A, Ham HR et al. Clinical usefulness of ultrashort-lived iridium- 191m from carbon-based generator system for the evaluation of the left ventricular function. J Nucl Med 1989; 30: 1025-31. 9. Brihaye C, Dewez S, Guillaume M et al. Reactor production and purification of osmium- 191 for use in a new OS-191/Ir-191m radionuclide generator system. Appl Radiat Isot 1989; 40: 183-9. 10. Performance standards of scintillation cameras, Standards Publication/No. NU 1-1986. National Electrical Manufacturers Association. 11. Ullman V, Husak V, Dubroka L. Deadtime correction in dynamic radionuclides studies by computer. Eur J Nucl Med 1978; 3: 197-202. 12. Johnston AS, Arnold JE, Pinsky SM. Anger camera deadtime: marker source correction and two parameter model. J Nucl Med 1975; 16: 539. 13. Lange D, Hermann HJ, Wetzel E, Schenck P. Critical parameters to estimate the use of a scintillation camera in high dose dynamic studies. Medical Radionuclide Imaging (Proc. Symp. Los Angeles) 1. Vienna: IAEA 1977; 85-100. 14. Johnston AS, Gergans GA, Kim I et al. Deadtime of computers coupled with anger cameras: counting losses and false counts. Single photon emission computed tomography and other selected computer topics (Proc. Symp. Miami 1980). Sorenson, ed. New York: Society of Nuclear Medicine. 15. Dymond DS, Elliot A, Stone D et al. Factors that affect the reproducibility of measurements of left ventricular function from first pass radionculide ventriculograms. Circulation 1982; 65: 311-22. 19

20 2.2 Variability of left ventricular ejection fraction and volumes by quantitative gated SPET : influence of algorithm, pixel size and reconstruction parameters in normal and small-sized hearts. Anne-Sophie Hambye 1, Ann Vervaet 2, André Dobbeleir 2,3 1 Nuclear Medicine, CHU-Tivoli, La Louvière, Belgium 2 Nuclear Medicine, Middelheim Hospital, Antwerp, Belgium 3 Nuclear Medicine, University Hospital Ghent, Ghent, Belgium Eur J Nucl Med Mol Imaging 2004; 31: 1606-1613. Abstract Several software are commercially available for quantification of left ventricle ejection fraction and volumes from myocardial gated SPET, all with a high reproducibility. However, their accuracy has been questioned in patients with a small-sized heart. This study aimed at evaluating the performances of different software and the influence of modifications in acquisition or reconstruction parameters on ejection fraction and volumes measurements, depending on the heart size. Methods: Sixty-four 2 and 128 2 matrix size acquisitions were consecutively obtained in 31 patients referred for gated SPET. After reconstruction by filtered backprojection (Butterworth, 0.4, 0.5 or 0.6 cyc/cm cutoff, order 6), LVEF and volumes were computed with different software (3 versions of Quantitative Gated SPECT (QGS), Emory Cardiac Toolbox (ECT) and the Stanford University (SU) Medical School algorithm), and processing workstations. Depending upon their end-systolic volume (ESV), patients were classified into 2 groups: Group I (ESV>30ml, n=14) and Group II (ESV <30ml, n=17). Agreement between the different software, and the influence of matrix size and sharpness of the filter on LVEF and volumes were evaluated in both groups. Results: In Group I, the correlation coefficients between the different methods ranged from 0.82 to 0.94 except for SU (r=0.77), and were slightly lower for volumes than ejection fraction. Mean differences between the methods were not significant, except for ECT which LVEF values were systematically higher by more than 10%. Changes in matrix size had no significant influence on LVEF or volumes. On the other hand, a sharper filter was associated with significantly larger volume values though this did usually not result in significant LVEF changes. In Group II, many patients had a LVEF at the higher range. The correlations coefficients between the different methods ranged between 0.80 and 0.96 except for SU (r=0.49), and were slightly worse for volumes than LVEF values. Contrary to Group I, a majority of mean differences between LVEF measurements was significant. LVEF was systematically the highest by ECT and the lowest by SU. With QGS, changes in matrix size from 64 2 to 128 2 were associated with significantly larger volumes as well as lower LVEF values. Increasing the filter cutoff frequency had the same effect. With SU-Segami, a larger matrix was associated with larger end-diastolic and smaller end-systolic volumes, resulting in a highly significant increase in LVEF. Increasing the filter sharpness on the other hand had no influence on LVEF though the measured volumes were significantly larger. Conclusion: In patients with a normal-sized heart, LVEF and volume estimates computed from different commercially available software for quantitative gated SPET are well correlated. LVEF and volumes are little sensitive to changes in matrix size. Smoothing on the other hand was associated with significant changes in volumes but usually not in LVEF values. However, owing to

the specific characteristics of each algorithm, software should not be interchanged for follow-up in an individual patient. In small-sized hearts on the other hand, both the used software and the matrix size or smoothing significantly influence the results of quantitative gated SPET. LVEF at the higher range are frequently observed with all the studied software except for SU-Segami. A larger matrix or a sharper filter could be suggested to enhance the accuracy of most commercial software, more particularly in patients with a small heart. Keywords: quantitative gated SPET LVEF small heart inter-software comparison Introduction Gated myocardial SPET has become the state-of-the-art for myocardial perfusion imaging, offering the simultaneous evaluation of left ventricular perfusion and function with a single test. Different methods to quantify left ventricular ejection fraction (LVEF) and volumes have been described [1-6], all with a high reproducibility and a good agreement with various non nuclear or nuclear techniques [6-10]. However, owing to the specific characteristics of each algorithm, software interchangeability for repeated examinations in an individual patient should not be recommended [9,11] despite the good correlations reported between different software computing the same gated SPET data [8,9,11]. Moreover, experimental data have revealed the sensitivity of gated SPET measured LVEF to particular acquisition conditions such as time of imaging, background activity or injected dose [12], filtering and zooming [13-15], and larger discrepancies between the methods have been described for LV volumes [8], particularly at both ends of the scope of volume values. Another problem in using quantitative gated SPET for LVEF calculation is encountered in patients with a small heart such as children or some small women. Indeed, due to the limited spatial resolution of the gamma cameras, the opposite endocardial edges of the left ventricle overlap, so that the ventricular cavity may become almost virtual especially at end-systole. This results in an underestimation of volumes, hence overestimation of LVEF [13-17], particularly using algorithms based upon edge detection. The purpose of our study was to compare LVEF and volumes computed from the same gated SPET data by different versions of the QGS-package [1], the Emory Cardiac Toolbox [4,5] and the Standford-University algorithm [6], and to evaluate the influence of filter and matrix size on the measurements. Material and methods Patients and acquisition During a 3-month period, QGS-analysis [1] was systematically performed in all patients undergoing a stress test as a part of a two-day stress-rest gated myocardial SPET. Depending upon their endsystolic volume (ESV) calculated on a GE-Elscint Expert system, the patients were classified into a group with a normal or large-sized heart (ESV >30 ml, Group I) and a group with a small-sized heart (ESV <30 ml, Group II). This value of 30ml-ESV was chosen based upon data from Ford et al, reporting that the difference between measured and true LVEF in a cardiac phantom becomes pronounced when the end-diastolic volume is <70 ml and the true LVEF is >40% [14]. Clinical characteristics of the both patients groups are reported in Table 1. Among those who required a comparative rest test, 31 underwent two consecutive gated SPET at rest: 14 of Group I and 17 of Group II. Decision to perform this double rest study was based solely upon the availability of free time-slots on the gamma-camera. The first acquisition in matrix 64 2, zoom 1.28 (6.9 mm-pixel size) started about 1 hour after injection of 740-1000 MBq 99mTcsestamibi and was immediately followed by a second acquisition in a 128 2 matrix, zoom 1.28 (3.45 mm-pixel size). Both SPET acquisitions lasted 25-30 minutes and were performed with a GE- 21

22 Elscint VariCam dual-head gamma camera equipped with VPC-35 collimators (system resolution of 9.0 mm FWHM at 10 cm distance; 290 cpm/µci), using an eight-bin gated protocol (90 projections (45/head) of 35 second/each; 360 -rotation; automatic body-contouring). Gated SPET Analysis Acquisition data sets were transferred from the GE-Elscint Expert system to a Sun Ultra10 Link Medical system (Link Medical, Hamshire, UK), a PC-Windows NT GE system (GE Medical Systems, Milwaukee, USA) and a PC-Windows NT Segami system (Segami, Columbia, USA) using DicomP10, and to a Nuclear Diagnostic Hermes system (Nuclear Diagnostic, Stockholm, Sweden) using modified interfile. The rough 64 2 and 128 2 matrix gated SPET acquisitions were reconstructed with Butterworth filters of 0.4, 0.5 or 0.6 cyc/cm cutoff (order 6) on different workstations. LVEF and volumes were automatically quantified from the gated coronal slices using commercially available software routinely used by the nuclear medicine community (three versions of Quantitative Gated SPECT (QGS), Cedars-Sinai Medical Center, Los Angeles, CA; Emory Cardiac Toolbox (ECT), Emory University, Atlanta, GA; Stanford University (SU) Medical School algorithm). These six different processings will be further referred to as QGS-Link, QGS-GE, QGS- Hermes, QGS-eNTEGRA, ECT-eNTEGRA and SU-Segami respectively. All have been described in detail elsewehere [1,3-5] and widely validated. Table 1. Clinical characteristics of the patient population (p=ns if >0.05). Group I: ESV > 30ml; Group II: ESV <30 ml; CRF: cardiovascular risk factors; MI: myocardial infarction; bicycle: upright bicycle stress test, 25W increment/2 min up to maximum heart rate; adenosine: 140µg/kg.min during 6 minutes; dobutamine: 10 to 40µg/kg.min with 3-min increments, + atropine if required. Group I (n=14) Group II (n=17) P value Age (years); mean±sd 55.3±14.6 65.1±12.1 0.048 Gender (M/F) 7 / 7 2 / 15 0.044 CRF 7 10 NS Prior MI 5 1 NS Prior revascularization 5 4 NS Referral reason Chest pain Abnormal stress EKG Other Kind of stress test (bicycle/adenosine/dobutamine) 9 / 4 / 1 8 / 9 / 0 9 2 3 16 1 0 NS NS NS NS Evidence of stress ischemia on SPECT 5 6 NS

23 Statistical analysis Results are expressed in absolute EF units for LVEF and in ml for the volumes. All statistical analyses were performed using the SPSS statistical program package (SPSS Inc, Chicago, USA). Inter-method variability was expressed as mean difference +/- SD. The significance of the difference between two groups of data was assessed by the paired or unpaired Student s t-test and chi squared test or Fisher s exact test, when appropriate. Paired data among three or more groups were compared using repeated measurements ANOVA. A p value of 0.05 or less was considered significant. To identify the differences for multiple testing, a Bonferroni correction was applied for comparing each pair of methods. With this correction, a p value of less than 0.0033 was considered significant. Pearson correlation coefficients were calculated, and Bland-Altman plots [18] were generated to search for trends by plotting the differences versus averages of paired values. For this part of the analysis, QGS-Link was arbitrarily chosen as a reference against which the other methods were plotted, as it constituted the last version of the most widely spread quantification method. Fig 1. Bland-Altman plots showing the agreement for ejection fraction between the reference method (QGS- Link) and the other packages. (ESV>30ml in open circles; ESV<30ml in solid circles). LVEF: left ventricular ejection fraction; ESV: end-systolic volume. QGS GE - QGS Link 35 30 25 20 15 10 5 0-50,0 20,0 40,0 60,0 80,0 100,0-10 -15-20 -25 mean QGS Hermes-QGS Link 35 30 25 20 15 10 5 0-50,0 20,0 40,0 60,0 80,0 100,0-10 -15-20 -25 mean QGS entegra-qgs Link 35 30 25 20 15 10 5 0-50,0 20,0 40,0 60,0 80,0 100,0-10 -15-20 -25 mean SU Segami-QGS Link 35 30 25 20 15 10 5 0-50,0 20,0 40,0 60,0 80,0 100,0-10 -15-20 -25 mean ECT entegra-qgs Link 35 30 25 20 15 10 5 0-50,0 20,0 40,0 60,0 80,0 100,0-10 -15-20 -25 mean

24 Results Using repeated measures analysis of variance between the six processings and the two groups of patients, a highly significant interaction was found, indicating that the impact of the software differed for the two patients groups. In addition, a significant overall difference was found between the six methods and also between the two patients groups (both p<0.001). Influence of the processing algorithm on ejection fraction and volume values. Group I (ESV> 30 ml). The different methods were fairly correlated with QGS-Link, with r values ranging between 0.82 and 0.94, except for SU-Segami (r=0.77). Mean LVEF and volume values were quite similar for the different methods, except for ECTeNTEGRA which resulted in higher LVEF (Table 2). By Bland-Altman analysis, no significant trend toward higher or lower LVEF was found across the whole range of values for any method but ECT-eNTEGRA compared to QGS-Link (Figure 1, open circles). By paired Student s t-test, highly significant differences (0.0001<p<0.0033) were noted between ECT-eNTEGRA and the other methods for LVEF and end-systolic but not for end-diastolic volume values (Table 3). Significant differences were also found for volume values between the QGS versions. Group II (ESV<30 ml) In keeping with Group I, all methods except SU-Segami correlated well with QGS-Link (r values between 0.80 and 0.96; r=0.49 for SU-Segami). Mean LVEF values were above 70% for all programs except for SU-Segami (mean LVEF: 60.4%, Table 2), and was highest by ECT-eNTEGRA. Opposite to Group I however, inter-method variability was quite large and most mean LVEF differences were significant (Table 3). Significant disparities in volume estimates were more frequent for end-systolic than end-diastolic volumes, and were particularly large for SU-Segami (between 10ml and 20ml, all p values <0.0001, Table 3). Compared to QGS-Link, LVEF was systematically higher by ECT-eNTEGRA (8.1±5.46%) and lower by SU-Segami (-13.7±8.01%) as shown on the Bland-Altman plots (Figure 1, solid circles). More surprisingly, a small but systematic difference in LVEF was also found by Bland-Altman analysis between the three versions of the QGS software, reaching the level of statistical significance for QGS-Hermes (Table 3). Table 2. Mean±SD ejection fraction and volumes for the different software (Group I: ESV > 30ml; Group II: ESV <30 ml; LVEF: left ventricular ejection fraction; EDV: end-diastolic volume; ESV: end-systolic volume). QGS Link QGS GE QGS Hermes QGS entegra ECT entegra SU Segami Group I LVEF (%) 45.1±12.98 47.4±12.43 49.4±12.48 47.5±13.53 62.0±14.13 50.1±12.67 EDV (ml) 119.5±66.24 122.4±63.85 112.6±63.85 119.3±65.76 108.2±51.22 119.3±49.19 ESV (ml) 72.3±59.96 69.9±58.32 62.7±52.47 69.9±57.89 46.1±41.14 64.3±45.45 Group II LVEF (%) 74.5±9.06 70.1±7.35 78.1±8.49 73.1±7.80 82.4±8.24 60.4±5.43 EDV (ml) 53.6±17.23 57.8±15.68 51.6±18.21 55.6±17.31 54.9±17.25 70.9±15.25 ESV (ml) 14.8±9.02 17.1±6.86 12.8±8.38 15.9±8.41 10.1±5.76 28.5±7.65

25 Influence of filtering on left ventricular ejection fraction and volume values. For this part of the study, 64 2 matrix size images were used. Due to technical limitations of some programs at our disposal at the time of the study, only QGS-GE, QGS-Hermes and SU-Segami were compared. Group I (ESV> 30 ml). Increasing the cutoff frequency of the Butterworth filter from 0.4 to 0.6 cyc/cm (order 6) resulted in significantly larger volumes for both QGS versions, and smaller volumes for SU-Segami. However, the subsequent changes in LVEF were significant only by QGS-GE (Table 4). The effect of filtering was more striking for the end-systolic volumes in relative values, although the absolute changes were usually higher for the end-diastolic (Table 4). Group II (ESV<30 ml) Sharper filtering resulted in significantly larger volumes for QGS, and particularly the end-diastolic, and in smaller volumes for SU-Segami. The ensuing LVEF change was however significant only for the QGS-versions, the SU-Segami LVEF remaining remarkably stable (Table 4). Table 3. Mean±SD difference in ejection fraction and volumes value according to the processing method used. (Group I: ESV > 30ml; Group II: ESV <30 ml; LVEF: left ventricular ejection fraction (%); EDV: enddiastolic volume (ml); ESV: end-systolic volume (ml)). P values are calculated using the Student s paired t-test after Bonferroni correction. *: 0.0001<p<0.0033; **: p<0.0001. All p values >0.0033 are considered as not significant. Group I Group II LVEF EDV ESV LVEF EDV ESV QGS Link-QGS GE -2.2+/-7.61-2.9+/-12.26 2.4+/-11.52 3.3+/-4.39-3.8+/-6.81-1.6+/-3.07 QGS Link-QGS Hermes -4.3+/-4.60 6.9+/-9.33 9.6+/-8.36 * -2.9+/-2.50 * 2.2+/-8.27 1.7+/-2.95 QGS Link-QGS enteg -2.3+/-7.57 1.8+/-14.59 3.8+/-13.58 1.1+/-2.47-0.9+/-2.76-0.7+/-1.88 QGS Link-ECT enteg -16.9+/-8.16 ** 11.3+/-21.95 26.1+/-22.52 * -8.1+/-5.46 ** 0.6+/-8.79 5.1+/-4.50 * QGS Link-SU Segami -4.9+/-8.69 0.2+/-23.9 8.0+/-19.48 13.7+/-8.01 ** -17.5+/-6.09 ** -13.5+/-4.29 ** QGS GE-QGS Hermes -2.1+/-5.94 9.8+/-8.46 * 7.1+/-10.60-6.2+/-3.42 ** 5.6+/-11.9 3.1+/-3.29 QGS GE-QGS enteg -1.0+/-3.74 5.4+/-10.53 2.5+/-8.56-2.4+/-3.05 2.8+/-6.88 1.1+/-2.82 QGS GE-ECT enteg -14.6+/-4.8 ** 14.2+/-17.68 23.7+/-18.94 * -11.8+/-6.35 ** 3.5+/-11.91 6.9+/-2.53 ** QGS GE-SU Segami -2.7+/-4.91 3.1+/-18.03 5.6+/-15.16 9.7+/-7.28 ** -13.2+/-7.61 ** -11.4+/-4.65 ** QGS ent-qgs Herm -2.0+/-5.87 5.0+/-12.35 5.9+/-11.71-4.1+/-2.25 ** 2.9+/-7.65 2.4+/-1.91 * QGS ent-ect enteg -14.4+/-4.87 ** 10.9+/-16.8 23.0+/-16.84 * -9.4+/-5.32 ** 0.7+/-8.20 5.8+/-3.87 ** QGS enteg-su Seg -2.2+/-7.06-0.7+/-18.5 4.5+/-15.42 12.3+/-7.45 ** -16.4+/-4.92 ** -12.8+/-3.69 ** QGS Herm-ECT ent -12.6+/-7.25 ** 4.4+/-15.69 16.6+/-15.66 * -5.2+/-5.54-1.4+/-8.01 3.5+/-4.16 QGS Herm-SU Segami -0.6+/-6.77-6.6+/-16.29-1.6+/-11.98 17.1+/-8.43 ** -19.1+/-6.76 ** -15.1+/-3.92 ** SU Sega-ECT enteg -11.9+/-6.63 ** 11.1+/-12.45 18.1+/-9.4 ** -21.7+/-9.02 ** 17.1+/-8.27 ** 18.6+/-4.27 **

26 Table 4: Influence of the filter cutoff frequency (order 6) on mean values for ejection fraction, end-diastolic and end-systolic volumes. The p values are calculated by overall repeated measurements ANOVA (p=ns if >0.05) (Group I: ESV > 30ml; Group II: ESV <30 ml; BW: Butterworth filter; LVEF: left ventricular ejection fraction; EDV: end-diastolic volume; ESV: end-systolic volume). Group I Group II BW 0.4 BW 0.5 BW 0.6 P value BW 0.4 BW 0.5 BW 0.6 P value QGS GE QGS Hermes SU Segami EF (%) 48.1 47.4 45.1 0.008 72.6 70.1 69.8 0.003 EDV (ml) 111.1 122.4 122.3 <0.001 50.7 57.8 57.5 <0.0001 ESV (ml) 62.0 69.9 72.7 <0.0001 13.7 17.1 17.1 <0.0001 EF (%) 49.6 49.5 48.7 NS 82.5 78.1 77.1 0.007 EDV (ml) 100.9 112.6 118.9 <0.0001 47.8 51.6 56.3 <0.0001 ESV (ml) 55.7 62.7 67.7 0.002 10.0 12.8 13.9 0.004 EF (%) 49.5 50.1 49.0 NS 60.3 60.4 60.5 NS EDV (ml) 128.6 119.3 112.9 <0.0001 74.3 70.9 67.9 0.002 ESV (ml) 69.8 64.3 62.1 <0.001 29.8 28.5 26.9 0.003 Influence of matrix size on left ventricular ejection fraction and volume values. For this part of the study, the 0.5 cyc/cm Butterworth filter images (order 6) were processed by QGS-GE, QGS-Hermes and SU-Segami. Group I (ESV> 30 ml). In this group, modifying the matrix size did not significantly influence mean LVEF and volume values except for the end-diastolic volumes by SU-Segami (Table 5). Mean± SD differences (matrix 64 2 128 2 ) for LVEF, EDV and ESV were respectively 0.7±5.88%, 3.6±13.3 ml and 1.9±12.05 ml for QGS-GE, 0.6±6.15%, 1.7±8.97 ml and -1.3±8.96 ml for QGS-Hermes, and -3.9±7.18%, - 9.1±10.51 ml and -1.1±9.36 ml for SU-Segami. Group II (ESV<30 ml) Decreasing the pixel size from 6.9 to 3.45 mm significantly modified the LVEF and volume values regardless of the used processing (Table 5). Using QGS, a smaller pixel size was associated with lower LVEF and larger volumes. Mean± SD differences (matrix 64 2 128 2 ) for LVEF, EDV and ESV were respectively 2.8±4.36% (p=0.021), - 6.8±8.28 ml (p=0.004) and -4.1±4.24 ml (p=0.001) for QGS-GE, and 5.1±4.66% (p=0.001), 6.7±7.02 ml (p=0.003) and -3.8±3.07 ml (p<0.0001) for QGS-Hermes. The effect of a smaller pixel size seemed particularly marked for end-diastolic volumes of 60ml and below. Using SU-Segami, results were divergent for end-diastolic and end-systolic volumes, the former increasing from a mean value of 70.9 ml to 76.1ml for the 128 2 matrix (p=0.014), and the latter decreasing from 28.5 ml to 20.5 ml (p<0.001). As a consequence, LVEF increased by 12.9±5.74% on average (p<0.0001).

27 Table 5: Influence of the acquisition matrix (64 2 or 128 2 ) on mean values for ejection fraction, end-diastolic and end-systolic volumes. The p values are calculated by paired Student s t-test (p=ns if >0.05). (Group I: ESV > 30ml; Group II: ESV <30 ml; LVEF: left ventricular ejection fraction; EDV: end-diastolic volume; ESV: end-systolic volume). Group I Group II Matrix size 64 2 128 2 P value 64 2 128 2 P value QGS GE QGS Hermes SU Segami LVEF (%) 47.4 46.6 NS 70.1 67.4 0.021 EDV (ml) 122.4 118.9 NS 57.8 64.6 0.004 ESV(ml) 69.9 68.0 NS 17.1 21.2 <0.001 LVEF (%) 49.4 48.9 NS 78.1 71.5 0.001 EDV (ml) 112.6 114.4 NS 51.6 58.5 0.003 ESV(ml) 62.7 64.0 NS 12.8 17.4 <0.0001 LVEF (%) 50.1 53.9 NS 60.5 73.4 <0.0001 EDV(ml) 119.3 128.4 0.006 70.9 76.1 0.014 ESV(ml) 64.3 65.4 NS 28.5 20.5 <0.0001 Discussion Using gated myocardial SPET, several algorithms have been developed for the calculation of LVEF and volumes, each owing its specific assumptions for left ventricle modeling. Among the various commercial programs, Cedars-Sinai Quantitative Gated SPECT (QGS, 1) is currently the most widely used in the clinical setting. Its reliability and reproducibility are excellent and have been validated against a whole range of methods. Nevertheless, with increased routine use, some limitations have appeared, such as a falsely elevated LVEF in patients with a small-sized heart like children or some women [14,16]. In patients with a normal- or large-sized heart, our study confirms the good agreement for LVEF between different processing methods [7-9] and the absence of significant bias through the whole range of LVEF values. Indeed, except for ECT-eNTEGRA that systematically overestimated LVEF by more than 10%, no significant method-related mean differences in LVEF were noted. This overestimation of LVEF by ECT has also been reported by others, including the authors of the program themselves [9,19], and might be due to specificities in time sampling or shape used for LV modeling [19]. Despite this good agreement, interchanging algorithms, or even consecutive versions of the same algorithm for follow-up studies in an individual patient should not be recommended because of the rather large standard deviation of the differences between the methods. For the volume values, and more particularly the end-systolic, we found a larger variability than for LVEF, with significant differences not only between ECT-eNTEGRA and the other programs, but also between the three versions of QGS, maybe due to (minor) modifications of its algorithm. In this patient population, increasing the matrix size had no significant influence on volume or LVEF values. Increasing the filter cutoff frequency on the other hand significantly modified the volume measurements, though this resulted in significant changes in LVEF only with QGS-GE. In patients with a small-sized heart, most mean differences in LVEF were significant despite a good agreement between the different methods except for SU-Segami. Moreover, a systematic bias was noted not only for ECT-eNTEGRA but also for SU-Segami which volumes were systematically

28 markedly larger, probably due to differences in the location of the ventricular wall which corresponds to the average position for the latter and to the endocardial surface for the former [7]. Also changes in matrix of filter cutoff significantly influenced volume and LVEF values in small hearts. This influence of the pixel size has already been reported by Nakajima et al in a cardiac phantom study [13]. They found a decrease from 49% to 3% in the overestimation of a 37-ml chamber volume by increasing the zoom from none to 2x during the acquisition, and confirmed their findings in a pediatric population, but only in children younger than 7 years [13]. However, the 1.28x-zooming applied in the present study is the maximum magnifying factor that can be used for a 60cm-field of view gamma camera without mechanical device keeping the heart in the center of rotation, so that we were compelled to increase the matrix from 64 2 to 128 2 to reduce the pixel size from 6.9 to 3.45 mm and so improve the delineation of the left ventricle endocardial border. Using QGS, this modification resulted in significantly larger volumes and lower LVEF, particularly for end-diastolic volumes of 60ml and below. By SU-Segami on the other hand, the combination of larger end-diastolic and smaller end-systolic volumes for a 128 2 matrix resulted in a highly significant increase in LVEF, probably because of an insufficient count density and thus enhanced statistical fluctuations. By increasing the cutoff frequency of the Butterworth filter from a smooth 0.4 to a sharper 0.6 cyc/cm, larger volumes and a significant decrease in LVEF was obtained by QGS. By SU-Segami on the contrary, LVEF remained stable despite significantly smaller volumes with a sharper filter, probably because of parallel changes in end-diastolic and end-systolic volumes. The influence of smoothing on LVEF and volumes could be due to the fact that, because of the limited spatial resolution of a gamma-camera, the proportion of LV volume contained in an individual pixel is larger in small than in large-sized hearts. In this way, changes in count density of the (especially endocardial) pixels related to the cardiac motion are probably more abrupt for higher cutoff frequency filtering. With a smooth filter, the systolo-diastolic transition in count density might be softer, hence volume estimates smaller and LVEF higher. The lesser filter-dependence observed with SU-Segami could be explained by the fact that its algorithm relies on the average ventricular wall position instead of the endocardial surface. This study compared different processing methods for quantitative estimates of LVEF and volumes using gated myocardial perfusion SPET. Despite good correlations with regard to the calculated values, clear differences were found between the algorithms, and more particularly between SU- Segami and the other methods, especially in patients with a small heart. No single external standard was available in our patients to determine the true values, so that the most recent version of the most widely used program was arbitrarily chosen as a reference. Therefore, the calculated results might be only a rough estimation of the patients real LVEF and volumes. However, since we aimed at correlating different processing methods computing the same gated SPET data, the use of an external standard does not seem an absolute prerequisite to validate the results. Another limitation consists in the use of low-energy general-purpose collimators (system resolution: 9.0 mm FWHM at 10 cm distance) for the gated SPET acquisition. Indeed, a high-resolution collimator should be preferred from a theoretical point of view since resolution recovery is expected to affect small volumes more than large. With a high resolution collimator, a 1.5 mm gain in resolution could be anticipated, but at the expense of a 40 %-count reduction which would require a smoother filter, hence loss of resolution, for acceptable image quality. The choice of general-purpose collimators constitutes thus a compromise between resolution and noise, especially using an automatic bodycontouring to reduce the patient-collimator distance. A last limitation concerns the small number of patients included. Despite this small sampling, highly significant results could be found so that this should not considered a major drawback, all the more as our purpose was to compare the different software currently available and not to identify the best of them.

29 Conclusion In patients with a normal-sized heart, quantitative estimates of left ventricular functional data computed from gated myocardial perfusion SPET by different commercially available software show excellent correlation. Inter-software, or even inter-version variability for an individual software is however present, especially regarding the volume values. Technical parameters such as matrix size or filter cutoff frequency have little influence on LVEF measurements but a sharper filter significantly modify the calculated volumes. Consequently, definition of specific normal limits should be advised for each algorithm, and software permutation should be avoided for follow-up studies in an individual patient. In small-sized hearts on the other hand, ejection fraction value in the (very) high range, most probably overestimated, is observed in a significant number of cases, so that the accuracy of gated SPET measured LVEF and volumes in these patients might be questioned. However, increasing the matrix size or the filter cutoff frequency results in significantly lower, probably more realistic LVEF with all the tested software except the SU-Segami. Although further confirmation of our results and validation of the correctness of the measurements is required, a smaller pixel size and/or a sharper filter might be suggested for quantitative gated SPET in patients with a small-sized heart. Acknowledgments The authors whish to thank H. Ham, MD, PhD, for his friendly comments and criticisms in the review of this manuscript. None of the authors has a financial interest in any software package. This study did not receive any vendor support. References 1. Germano G, Kiat H, Kavanagh PB, Moriel M, Mazzanti M, Su H, et al. Automatic quantification of ejection fraction from gated myocardial perfusion SPECT. J Nucl Med 1995; 36:2138-47. 2. Germano G, Kavanagh PB, Kavanagh JT, Wishner SH, Berman DS, Kavanagh GJ. Repeatability of automatic left ventricular cavity volume measurements from myocardial perfusion SPECT. J Nucl Cardiol 1998; 5:477-483. 3. Goris ML,Thompson C, Malone L, PR Franken. Modeling the integration of myocardial regional perfusion and function. Nucl Med Commun 1994; 15: 9-20. 4. Faber TL, Akers MS, Peshock RM, Corbett JR. Three dimensional motion and perfusion quantification in gated single-photon emission computed tomograms. J. Nucl Med 1991; 32:2311-2317. 5. Faber Tl, Cooke CD, Folks RD et al. Left ventricular function and perfusion from gated perfusion images : an integrated method. J Nucl Med 1999; 40: 650-659. 6. Nichols K, DePuey RG, Rozanski A. Automation of gated tomography left ventricular ejection fraction. J Nucl Cardiol 1996; 3: 475-482 7. Everaert H, Bossuyt A, Franken P. Left ventricular ejection fraction and volumes from gated single photon emission tomographic myocardial perfusion images: Comparison between two algorithms working in three-dimensional space. J Nuclear Cardiology 1997;4:472-476. 8. Nichols K, Lefkowitz D, Faber T, et al. Echocardiographic validation of gated SPECT ventricular function measurements. J Nucl Med 2000;41:1308-14.

30 9. Nakajima J, Higuchi T, Taki J, Kawano M, Tonami N. Accuracy of ventricular volume and ejection fraction measured by gated myocardial SPECT: Comparison of 4 software Programs. J Nucl Med 2001; 42: 1571-78. 10. Vourvouri E, Poldermans D, Bax JJ, et al. Evaluation of left ventricular function and volumes in patients with ischaemic cardiomyopathy : gated single-photon emission computed tomography versus two-dimentional echocardiography. Eur J Nucl Med 2001;28:1610-15. 11. Lum DP, Coel MN. Comparison of automatic quantification software for the measurement of ventricular volume and ejection fraction in gated myocardial perfusion SPECT. Nucl Med Comm 2003; 24: 259-266 12. Vallejo E, Dione DP, Bruni WL, et al. Reproducibility and accuracy of gated SPECT for determination of left ventricular volume and ejection fraction: experimental validation using MRI. J Nucl Med 2000; 41:874-882. 13. Nakajima K, Taki J, Higuchi T, Kawano M, Taniguchi M, Maruhashi K, Sakazume S, Tonami N. Gated SPET quantification of small hearts: mathematical simulation and clinical application. Eur J Nucl Med 2000;27:1372-79. 14. Ford P, Chatziioannou S, Moore H, Dhekne R. Overestimation of the LVEF by quantitative gated SPECT in simulated left ventricles. J Nucl Med 2001;42:454-459. 15. Manrique A, Hitzel a, Gardin I, Dacher JN, Vera P. Influence of Wiener filter in determining the left ventricle volume and ejection fraction using thallium-201 gated SPECT. Nucl Med Comm 2003; 24: 907-914 16. De Bondt P, Van de Wiele C, De Sutter J, De Winter F, De Backer G, Dierckx RA. Ageand gender-specific differences in left ventricular cardiac function and volumes determined by gated SPET. Eur J Nucl Med 2001;28:620-24. 17. Achtert AD, King MA, Darlberg ST, et al. An investigation of the estimation of ejection fractions and cardiac volumes by a quantitative gated SPECT software package in simulated gated SPECT images. J Nucl Cardiol 1998;5:144-152. 18. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurements. Lancet 1986; 1: 307-310. 19. Nichols K, Santana CA, Folks R et al. Comparison between ECT and QGS for assessment of left ventricular function from gated myocardial perfusion SPECT. J Nucl Cardiol 2002; 9:285-93