- Email: [email protected]
Rapid and accurate noninvasive assessment of global left ventricular systolic function using biplane advanced automated contour tracking method
Rapid and Accurate Noninvasive Assessment of Global Left Ventricular Systolic Function Using Biplane Advanced Automated Contour Tracking Method Kenichi Sugioka, MD, Takeshi Hozumi, MD, Hiroyuki Watanabe, MD, Hiroyuki Yamagishi, MD, Yoshiki Matsumura, MD, Yasuhiko Takemoto, MD, Takashi Muro, MD, Minoru Yoshiyama, MD, Kazuhide Takeuchi, MD, and Junichi Yoshikawa, MD, Osaka, Japan
Background: The advanced automated contour tracking (AACT) method has been newly developed for automated detection of the left ventricular endocardial boundary. Left ventricular ejection fraction (LVEF) may be estimated by applying the AACT method to 2 orthogonal planes of patients even when regional wall-motion abnormalities exist. The purpose of this study was to examine the reliability of the biplane AACT method in the measurement of LVEF in patients with suggested ischemic heart disease with use of quantitative gated single photon emission computed tomography (QGS) as a reference standard. Methods: The study population consisted of 47 consecutive patients with suggested ischemic heart disease. All patients underwent 2-dimensional echocardiography and QGS. Biplane LVEF from apical 4- and 2-chamber views was measured offline by the AACT method using disk summation method. The accu-
Left ventricular (LV) ejection fraction (EF) is an
important prognostic predictor and physiologic index for patients with ischemic heart disease (IHD).1 For noninvasive quantitative analysis of LV volume and EF, disk summation method has been used by manual tracing of the LV endocardial border using 2-dimensional echocardiography.2-7 In this method, however, the process of manual tracing of the endocardium is tedious, time-consuming, and a cause of observer variability. Accordingly, several offline automated tracing methods for echocardioFrom the Department of Internal Medicine and Cardiology, Osaka City University Medical School, Osaka, Japan. Presented in part at the 52nd Annual Scientific Session of the American College of Cardiology, Chicago, Ill, April 1, 2003. Reprint requests: Takeshi Hozumi, MD, Department of Internal Medicine and Cardiology, Osaka City University Medical School, 1-4-3, Asahi-machi, Abeno-ku, Osaka, 545-8585 Japan (E-mail: [email protected]). Copyright 2003 by the American Society of Echocardiography. 0894-7317/2003/$30.00 ⫹ 0 doi:10.1067/j.echo.2003.08.014
racy of the AACT method for LVEF measurement was determined in comparison with QGS. Results: In 41 (29 with and 12 without regional wall-motion abnormalities) of 47 patients (87%), automated tracing of the endocardial border was adequately achieved with the AACT method. LVEF measured by the AACT method correlated well with that measured by QGS (y ⴝ 0.97x ⴙ2.4, r ⴝ 0.91). The mean difference between AACT and QGS was 0.6 ⴞ 5.5% (mean ⴞ SD). The mean time required for analysis of 1 set of images during 1 cardiac cycle by the AACT method was much shorter than that required by manual tracing method (7 ⴞ 1 vs 37 ⴞ 4 seconds, P < .0001). Conclusion: The biplane AACT method provides accurate and quick measurement of LVEF in patients even with regional wall-motion abnormalities. (J Am Soc Echocardiogr 2003;16:1237-43.)
graphic images have been tested to overcome these limitations.8-10 Previously, the automated contour tracking (ACT) method was introduced as an automated technique for LV volume and EF measurement.11,12 This method is on the basis of the automated endocardial border detection technique under the principle of weighted separatability of the image and contour smoothing with minimization of elastic energy.13-15 However, application of this method is limited to patients with high-quality echocardiographic images.11,12 Recently, the ACT method has been armed with a new technique, the “partial shape constant contour model,”16 to improve its accuracy. This new ACT method is called the “advanced ACT (AACT) method” and may allow the assessment of LV systolic function through automated LV border detection even in cases with suboptimal images. In addition, an advantage of this method is its capability of biplane assessment, which may enable accurate LV volume and EF assessment of patients even when regional wallmotion abnormalities exist. The purpose of this
1237
1238 Sugioka et al
Journal of the American Society of Echocardiography December 2003
Figure 1 Description of partial shape constraint contour model. Basic principles of pattern matching (A) and contour restraint (B). LV, Left ventricular.
study, therefore, was to examine the accuracy of the biplane LVEF derived by the AACT method in patients with suggested IHD with use of quantitative gated single photon emission computed tomography (SPECT) as a reference standard.
METHODS Study Patients The study population consisted of 47 consecutive patients (35 men and 12 women; age 29 to 79 years, mean 64 ⫾ 10) who were scheduled for quantitative gated SPECT (QGS) with in sinus rhythm and with suggested IHD. For each patient, 2-dimensional echocardiography and QGS were performed within 48 hours. There were no changes in any patient’s clinical status during this period. Informed written consent was obtained from all patients. The Principles of the AACT Method The AACT method is now commercially available in the newest ultrasound system (SSA-770A Aplio, Toshiba Corp, Tochigi, Japan). The technologic framework of the AACT method has 3 features: (1) weighted separability for edge detection of the endocardial border; (2) contour smoothing using minimization of elastic energy; and (3) partial shape constant contour model. The principles of weighed separability and contour smoothing were described previously.11-15 The newly developed “partial shape constant contour model” consists of 2 features, pattern matching
and contour constraint.16-18 In pattern matching, the intensity pattern of the mitral annulus is stored, and pattern matching against this stored intensity pattern is performed for each frame in the specified measurement range to monitor the position of the mitral annulus (Figure 1, A). This processing is expected to avoid misidentification of the contours because of the chordae tendineae and to improve capability of tracing of the endocardium that can not be clearly displayed. In contour constraint, the endocardial contour extraction range is restricted by 3 sample points, which are placed on both sides of the mitral annulus and the apex (Figure 1, B). This processing is expected to improve trace fitting in cases in which the apical endocardium cannot be clearly visualized. LV Volume and EF Measurement by the AACT Method Echocardiographic measurement of LV volume and EF were performed with an ultrasound system (SSA-770A Aplio, Toshiba Corp) using a 2.5-MHz (2-4 MHz) broadband phased-array transducer. Both apical 4- and 2-chamber views were obtained by means of 2-dimensional echocardiography using harmonic imaging. Clips of echocardiographic images of the LV in these views were acquired and stored digitally in the ultrasound system for offline analysis of LV volume and EF using the AACT method (37 frames/s). In each case, the operator was required to place 3 sample points on both sides of the mitral annulus and the LV apex, in apical views of the end-diastolic image selected with the R wave of the electrocardiogram. Then,
Journal of the American Society of Echocardiography Volume 16 Number 12
Sugioka et al 1239
Figure 2 Steps of automated contour tracking procedure in 4-chamber views: (1) operator marks 3 points at both sides of mitral annulus and at apex in apical views of end-diastolic image that selected with R wave of electrocardiogram; (2) contour of left ventricular (LV) endocardial border is automatically tracked; and (3) extraction of endocardial border of LV cavity is completed in every frame throughout 1 cardiac cycle.
Figure 3 Example of end-diastolic volume, end-systolic volume, and ejection fraction from apical 4- and 2-chamber views by advanced automated contour tracking method. LV, Left ventricular. the endocardial border was identified automatically, and extraction of the endocardial border of the LV cavity was completed in every frame throughout 1 cardiac cycle (Figure 2). The end-systolic frame was selected automatically by estimating smallest cavity volume. In this method, echocardiographic LV volume determined from 4- and 2-chamber apical views and biplane EF were calculated automatically, using disk summation method with a computer incorporated in this ultrasound system (Figure 3). All measurements were performed by a single investigator who was blinded to the gated SPECT results. Unsuccessful tracing was visually defined as cases with obvious discrepancies (ie, tracing inside or outside the LV) between the actual and detected endocardial border. A total of 25 patients were randomly selected for assessment of the intraobserver and interobserver variabilities of the AACT method and for measurement of study time required for analysis of LV volume and EF from an apical 4-chamber image in both the AACT method and manual tracing method. Images obtained in these patients were analyzed by 2 independent observers, without
knowledge of patient identity or previous results. In addition, intraobserver analyses were performed by the same observer 2 weeks after the first evaluation. Interobserver variability was calculated as the SD of the differences between the measurements of 2 independent observers and expressed as a percentage of the average value. Intraobserver variability was calculated as the SD of the differences between the first and second determination for a single observer and expressed as a percentage of the average value. QGS QGS was performed as previously described.19,20 All patients underwent thallium-201-gated SPECT after thallium–201 (111 and 74 MBq for patients without and with previous myocardial infarction, respectively) was injected intravenously. Initial images were obtained immediately after the termination of exercise, and delayed images were obtained 4 hours later. For patients with previous myocardial infarction, an additional dose of 37 MBq thallium–201 was injected at rest immediately after the acquisition of
1240 Sugioka et al
Journal of the American Society of Echocardiography December 2003
Figure 4 Scatter plots showing correlation between advanced automated contour tracking (AACT) method and quantitative gated single photon emission computed tomography (QGS) in measurement of end-diastolic volume (left) and mean end-diastolic volume by AACT and QGS (x axis) and difference between AACT and QGS (y axis) (right). Mean difference and 2 SD limits are shown. SEE, Standard error of the estimate.
delayed images, and the reinjection images were obtained 20 minutes later. SPECT was performed with a 2-detector gamma camera (Vertex, ADAC Laboratories, Milpitas, Calif) equipped with low-energy, general purpose collimators, with the detectors set to form a 90-degree angle. A total of 32 equidistant projections were acquired over 180 degrees in a 64 ⫻ 64 matrix from the 45-degree right anterior oblique projection to the 45-degree left posterior oblique projection with an energy window of 70 keV ⫾ 10%. Electrocardiogram-gated images were acquired with 40-second, 6-degree angular steps. At each projection, 8 frames/ cardiac cycle were acquired, with an allowable change in R-R interval of the mean value ⫾ 20%. Transaxial slices of 4.7-mm-pixel thickness were reconstructed using a Butterworth filter (order, 5.0; critical frequency, 0.35 cycle/ pixel) and the filtered backprojection method (ramp filter) on a processing computer (Pegasys, ADAC Laboratories) with an automatic reorientation algorithm for SPECT (AutoSPECT, Cedars-Sinai Medical Center, Los Angeles, Calif).21 No attenuation correction was applied. A completely automated software program that analyzes LV function (Quantitative Gated SPECT, Cedars-Sinai Medical Center) was used to calculate LV volume and global EF.22,23 Rest LV volume and EF were derived from delayed images for patients without reinjection images and from reinjection images for patients with reinjection images. Statistics Linear regression analysis was used to compare QGS and the AACT assessment of LV volumes in end-diastole and end-systole, and EF. Differences in the measurements obtained by each method are expressed as mean ⫾ SD. Analyses of the differences in measurements were performed using the technique of Bland and Altman.24 We used the paired t test to compare the time required for
LVEF calculation by the AACT method with that required by the manual method. P values ⬍ .05 were considered statistically significant.
RESULTS The diagnoses of the 47 patients were as follows: 14 patients with angina pectoris; 16 with myocardial infarction; 9 with both angina pectoris and myocardial infarction; 3 with dilated cardiomyopathy; and 5 healthy patients. In 41 of 47 patients (87%), automated tracing of the endocardial border was achieved adequately by the AACT method. Of these, there were 29 patients with regional wall-motion abnormalities and 12 without them. Comparison of LV Measurements Between QGS and the AACT Method LV end-diastolic volume measured by the AACT method demonstrated excellent correlation with that by QGS at end-diastole (r ⫽ 0.96, y ⫽ 0.85x ⫹2.4, standard error of the estimate ⫽ 4.2 mL). The mean difference in LV end-diastolic volume between AACT and QGS was 10.8 ⫾ 13.0 mL (Figure 4). LV end-systolic volume measured by the AACT method also demonstrated excellent correlation with that from QGS (r ⫽ 0.98, y ⫽ 0.87x ⫺4.0, standard error of the estimate ⫽ 1.8 mL). The mean difference in LV end-systolic volume measurement between AACT and QGS was 6.7 ⫾ 7.4 mL (Figure 5). LVEF measured by the AACT method correlated well with that by QGS (r ⫽ 0.91, y ⫽ 0.97x ⫹2.4, standard error of the estimate ⫽ 3.7%). The mean difference between AACT and QGS was 0.6 ⫾ 5.5% (Figure 6).
Journal of the American Society of Echocardiography Volume 16 Number 12
Sugioka et al 1241
Figure 5 Scatter plots showing correlation between advanced automated contour tracking (AACT) method and quantitative gated single photon emission computed tomography (QGS) in measurement of end-systolic volume (left) and mean end-systolic volume by AACT and QGS (x axis) and difference between AACT and QGS (y axis) (right). Mean difference and 2 SD limits are shown. SEE, Standard error of the estimate.
Figure 6 Scatter plots showing correlation between advanced automated contour tracking (AACT) method and quantitative gated single photon emission computed tomography (QGS) in measurement of ejection fraction (EF) (left) and mean EF by AACT and QGS (x axis) and difference between AACT and QGS (y axis) (right). Mean difference and 2 SD limits are shown. SEE, Standard error of the estimate.
Time Efficiency and Reproducibility The AACT method required 7 ⫾ 1 seconds for LV volume and EF measurement from 1 set of images during 1 cardiac cycle, whereas 37 ⫾ 4 seconds were needed for the manual tracing method. The study time was significantly shorter using the AACT method than that by manual tracing method (P ⬍ .0001). With the AACT method, intraobserver and interobserver variabilities for LVEF measurement were 4.5% and 7.3%, respectively.
DISCUSSION This is the first study of the biplane, automated assessment of LV volume and EF using the AACT method. In this study, we demonstrated that LVEF was accurately quantified using the AACT method
on patients even when regional wall-motion abnormalities existed. Furthermore, the time needed for LVEF measurement was much shorter using the AACT method than with the manual tracing method. Previous Echocardiographic Estimation of LVEF Two-dimensional echocardiography is a noninvasive and widely accepted method for the measurement of LV function in the clinical setting. LVEF measurement by manual tracing method using apical views, which is recommended by the American Society of Echocardiography,7 is more reliable than that by the M-mode method.2,5 However, clinical use of this method is limited because it requires tedious and time-consuming processes. To improve on these limitations of the manual tracing method, several methods for offline automated border detection
1242 Sugioka et al
have been attempted.8,9 Geiser et al10 reported the autonomous boundary detection for epicardium and endocardium. This method was, however, limited to analysis from the parasternal short-axis view and to cases with high-quality images and with normal LV wall motion. For other offline automated techniques, application in clinical use is also hampered by requirement of processing and much processing time.8,9 Previously, the ACT method was developed for offline automated contour tracing of the LV endocardial boundary without tracing of a region of interest.11-15 The advantages of this method were its simple manual procedure and short processing time. Furthermore, results obtained from this method correlate highly with those of the manual method for estimation of LV cavity areas11 and with left ventriculography for measurement of LVEF in clinical studies.12 However, endocardial border detection by the ACT method often failed because of misidentification of the apex or mitral annulus for patients with suboptimal images. Moreover, in previous studies using the ACT method, patients with wall-motion abnormalities were excluded and the usefulness of this method was not established for patients with IHD.11,12 Advantages of the AACT Method The AACT method especially addresses the limitations of the previous ACT method and has some advantages. First, reliable LVEF measurement by the AACT method can be performed with a high rate of success. During development of the AACT method, technical efforts were focused on improving the accuracy of tracing by using partial shape constant contour model.16-18 This study demonstrated that the success rate of the AACT method in LVEF measurement was 87%, which is high enough for use in the clinical setting. In addition to technologic advancement in the AACT method, the use of harmonic imaging, which allows improvement of LV endocardial border delineation,25 may also contribute to accurate LVEF assessment. Second, the AACT method permits accurate measurement of LVEF even for patients with regional wall-motion abnormalities, because it uses 2 orthogonal planes for biplane LVEF measurement. To our knowledge, no automated method other than the AACT method can measure LVEF in a biplane manner. As presented in this study, biplane LVEF measurements using the AACT method agreed well with those using QGS as the gold standard in patients even when regional wall-motion abnormalities existed. The acoustic quantification method is a widespread method for automated border detection with online analysis.26-28 Although it is a powerful technique that provides real-time LV function, it is applied only in single plane for the assessment of
Journal of the American Society of Echocardiography December 2003
global LV systolic function. In cases with regional LV wall-motion abnormalities, however, LVEF derived from single-plane acoustic quantification may be less accurate compared with that from biplane method.28 By contrast, this AACT method can be applied even to patients with regional wall-motion abnormalities because LV volume is calculated by biplane method in this new automated system. Thus, extended application for patients with IHD for global LV systolic function may be one of the advantages of this method compared acoustic quantification method. Finally, the time of analysis in the AACT method is much shorter than that in the previous ACT method because of an improved calculation algorithm and by recent advancement in computer technology. In this study, study time required for measuring LVEF by the AACT method from 1 apical view was one-fifth that by the manual method. This quick LVEF measurement by the AACT method would be an advantage of this method in the clinical setting. Study Limitations There were a few patients with LVEF less than 35% included in this study. Our data may need to be confirmed in larger patient populations. In this study, the detected boundaries were not compared with any independent technique of boundary detection other than subjective visual interpretation. The accuracy of boundary detection by this AACT method may need to be evaluated in further studies. In addition, in the AACT method, automated detection of endocardial borders can be performed not only in end-systole and end-diastole frames, but in every frame throughout 1 cardiac cycle. In this study, however, we showed only the results of endocardial tracing in end-diastolic and end-systolic frames. Regarding theses frames, we showed a good relationship between AACT and QGS for LV volume. Because the results of endocardial tracing by AACT of endocardial tracing in every frame through a whole cardiac cycle was not shown in this study, it has not been clarified how wall-motion abnormalities affect automated tracing of LV endocardium by the AACT method. Regarding this issue, further investigations are necessary. Finally, although the success rate of the AACT method in LVEF measurement was high, automated tracing of the endocardial border was unable to be achieved adequately in 6 patients for the following reasons: tracing outside the LV lateral wall in 4-chamber view or anterior wall in 2-chamber view at end-diastole in 3 patients; tracing inside the LV in 3 patients; and inadequate tracing in the apical segment in 1 patient. Conclusions The biplane AACT method provides accurate and quick measurement of LVEF for patients even when
Journal of the American Society of Echocardiography Volume 16 Number 12
regional wall-motion abnormalities exist. This newly developed method is reliable in noninvasive evaluation of global LV systolic function. We acknowledge the assistance of Naoki Yoneyama, BS, Hiroyuki Tsujino, MS, Ryoichi Kanda, MS, and Masahide Nishiura, MS, of Toshiba Corporation in technical advices for preparing this manuscript. REFERENCES 1. The Multicenter Postinfarction Research Group. Risk stratification and survival after myocardial infarction. N Engl J Med 1983;309:331-6. 2. Schiller NB, Acquatella H, Ports TA, Drew D, Goerke J, Ringertz H, et al. Left ventricular volume from paired biplane two-dimensional echocardiography. Circulation 1979;60:54755. 3. Folland ED, Parisi AF, Moynihan PF, Jones DR, Feldman CL, Tow DE. Assessment of left ventricular ejection fraction and volumes by real-time, two-dimensional echocardiography. Circulation 1979;60:760-6. 4. Carr KW, Engler RL, Forsythe JR, Johnson AD, Gosink B. Measurement of left ventricular ejection fraction by mechanical cross-sectional echocardiography. Circulation 1979;59: 1196-206. 5. Starling MR, Crawford MH, Sorensen SG, Levi B, Richards KL, O’Rourke RA. Comparative accuracy of apical biplane cross-sectional echocardiography and gated equilibrium radionuclide angiography for estimating left ventricular size and performance. Circulation 1981;63:1075-84. 6. Silverman NH, Ports TA, Snider AR, Schiller NB, Carlsson E, Heilbron DC. Determination of left ventricular volume in children: echocardiographic and angiographic comparisons. Circulation 1980;62:548-57. 7. Schiller NB, Shah PM, Crawford M, DeMaria A, Devereux R, Feigenbaum H, et al. Recommendations for quantitation of the left ventricle by two-dimensional echocardiography. J Am Soc Echocardiogr 1989;2:358-67. 8. Bosch JG, Savalle LH, Burken G, Reiber JH. Evaluation of a semiautomatic contour detection approach in sequences of short-axis two-dimensional echocardiographic images. J Am Soc Echocardiogr 1995;8:810-21. 9. Cootes TF, Taylor CJ, Cooper DH. Active shape models: their training and application. Comput Vis Image Underst 1995; 61:38-59. 10. Geiser EA, Wilson DC, Wang DX, Conetta DA, Murphy JD, Hutson AD. Autonomous epicardial and endocardial boundary detection in echocardiographic short-axis images. J Am Soc Echocardiogr 1998;11:338-48. 11. Hozumi T, Yoshida K, Yoshioka H, Yagi T, Akasaka T, Takagi T, et al. Echocardiographic estimation of left ventricular cavity area with a newly developed automated contour tracking method. J Am Soc Echocardiogr 1997;10:822-9. 12. Sugioka K, Hozumi T, Yagi T, Yamamuro A, Akasaka T, Takeuchi K, et al. Automated quantification of left ventricular function by the automated contour tracking method. Echocardiography 2003;20:313-8.
Sugioka et al 1243
13. Cipolla R, Blake A. The dynamic analysis of apparent contours. Proc 3 Int J Comput Vis 1990;i:616-23. 14. Menet S, Saint-Marc P, Medioni G. Active contour models: overview, implementation and applications. IEEE Proc Int Conf Syst Man Cybern 1990;6:194-9. 15. Kass M, Witkin A, Terzopoulos D. Snakes: active contour models. Int J Comput Vis 1988;1:321-31. 16. Nishiura M, Yuasa M, Watanabe M. Active contour extraction method using partial shape constraint contour model. Syst Comput Jpn 2000;31:20-8. 17. Nishiura M, Furukawa R, Tsujino H, Kanda R. Automated tracking of the mitral annulus using two-dimensional pattern matching technique: a validation study [abstract]. J Am Soc Echocardiogr 2001;14:469. 18. Tsujino H, Nishiura M, Furukawa R, Kanda R. Quantitative assessment of the mitral annular motion using a two-dimensional automated annular tracking method [abstract]. J Am Soc Echocardiogr 2001;14:478. 19. Yamagishi H, Shirai N, Yoshiyama M, Teragaki M, Akioka K, Takeuchi K, et al. Incremental value of left ventricular ejection fraction for detection of multivessel coronary artery disease in exercise 201-Tl gated myocardial perfusion imaging. J Nucl Med 2002;43:131-9. 20. Shirai N, Yamagishi H, Yoshiyama M, Teragaki M, Akioka K, Takeuchi K, et al. Incremental value of assessment of regional wall motion for detection of multivessel coronary artery disease in exercise 201-Tl gated myocardial perfusion imaging. J Nucl Med 2002;43:443-50. 21. Germano G, Kavanagh PB, Su H, Mazzanti M, Kiat H, Hachamovitch R, et al. Automatic reorientation of threedimensional, transaxial myocardial perfusion SPECT images. J Nucl Med 1995;36:1107-14. 22. Germano G, Kiat H, Kavanagh PB, Moriel M, Mazzanti M, Su HT, et al. Automated quantification of ejection fraction from gated myocardial perfusion SPECT. J Nucl Med 1995; 36:2138-47. 23. Germano G, Erel J, Lewin H, Kavanagh PB, Berman DS. Automated quantitation of regional myocardial wall motion and thickening from gated technetium-99m sestamibi myocardial perfusion single-photon emission computed tomography. J Am Coll Cardiol 1997;30:1360-7. 24. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;1:307-10. 25. Spencer KT, Bednarz J, Rafter PG, Korcarz C, Lang RM. Use of harmonic imaging without echocardiographic contrast to improve two-dimensional image quality. Am J Cardiol 1998; 82:794-9. 26. Vandenberg BF, Rath LS, Stuhlmuller P, Melton HE Jr, Skorton DJ. Estimation of left ventricular cavity area with an on-line, semiautomated echocardiographic edge detection system. Circulation 1992;86:159-66. 27. Perez JE, Waggoner AD, Barzilai B, Melton HE Jr, Miller JG, Sobel BE. On-line assessment of ventricular function by automatic boundary detection and ultrasonic backscatter imaging. J Am Coll Cardiol 1992;19:313-20. 28. Bednarz JE, Marcus RH, Lang RM. Technical guideline for performing automated border detection studies. J Am Soc Echocardiogr 1995;8:293-305.
Background: The advanced automated contour tracking (AACT) method has been newly developed for automated detection of the left ventricular endocardial boundary. Left ventricular ejection fraction (LVEF) may be estimated by applying the AACT method to 2 orthogonal planes of patients even when regional wall-motion abnormalities exist. The purpose of this study was to examine the reliability of the biplane AACT method in the measurement of LVEF in patients with suggested ischemic heart disease with use of quantitative gated single photon emission computed tomography (QGS) as a reference standard. Methods: The study population consisted of 47 consecutive patients with suggested ischemic heart disease. All patients underwent 2-dimensional echocardiography and QGS. Biplane LVEF from apical 4- and 2-chamber views was measured offline by the AACT method using disk summation method. The accu-
Left ventricular (LV) ejection fraction (EF) is an
important prognostic predictor and physiologic index for patients with ischemic heart disease (IHD).1 For noninvasive quantitative analysis of LV volume and EF, disk summation method has been used by manual tracing of the LV endocardial border using 2-dimensional echocardiography.2-7 In this method, however, the process of manual tracing of the endocardium is tedious, time-consuming, and a cause of observer variability. Accordingly, several offline automated tracing methods for echocardioFrom the Department of Internal Medicine and Cardiology, Osaka City University Medical School, Osaka, Japan. Presented in part at the 52nd Annual Scientific Session of the American College of Cardiology, Chicago, Ill, April 1, 2003. Reprint requests: Takeshi Hozumi, MD, Department of Internal Medicine and Cardiology, Osaka City University Medical School, 1-4-3, Asahi-machi, Abeno-ku, Osaka, 545-8585 Japan (E-mail: [email protected]). Copyright 2003 by the American Society of Echocardiography. 0894-7317/2003/$30.00 ⫹ 0 doi:10.1067/j.echo.2003.08.014
racy of the AACT method for LVEF measurement was determined in comparison with QGS. Results: In 41 (29 with and 12 without regional wall-motion abnormalities) of 47 patients (87%), automated tracing of the endocardial border was adequately achieved with the AACT method. LVEF measured by the AACT method correlated well with that measured by QGS (y ⴝ 0.97x ⴙ2.4, r ⴝ 0.91). The mean difference between AACT and QGS was 0.6 ⴞ 5.5% (mean ⴞ SD). The mean time required for analysis of 1 set of images during 1 cardiac cycle by the AACT method was much shorter than that required by manual tracing method (7 ⴞ 1 vs 37 ⴞ 4 seconds, P < .0001). Conclusion: The biplane AACT method provides accurate and quick measurement of LVEF in patients even with regional wall-motion abnormalities. (J Am Soc Echocardiogr 2003;16:1237-43.)
graphic images have been tested to overcome these limitations.8-10 Previously, the automated contour tracking (ACT) method was introduced as an automated technique for LV volume and EF measurement.11,12 This method is on the basis of the automated endocardial border detection technique under the principle of weighted separatability of the image and contour smoothing with minimization of elastic energy.13-15 However, application of this method is limited to patients with high-quality echocardiographic images.11,12 Recently, the ACT method has been armed with a new technique, the “partial shape constant contour model,”16 to improve its accuracy. This new ACT method is called the “advanced ACT (AACT) method” and may allow the assessment of LV systolic function through automated LV border detection even in cases with suboptimal images. In addition, an advantage of this method is its capability of biplane assessment, which may enable accurate LV volume and EF assessment of patients even when regional wallmotion abnormalities exist. The purpose of this
1237
1238 Sugioka et al
Journal of the American Society of Echocardiography December 2003
Figure 1 Description of partial shape constraint contour model. Basic principles of pattern matching (A) and contour restraint (B). LV, Left ventricular.
study, therefore, was to examine the accuracy of the biplane LVEF derived by the AACT method in patients with suggested IHD with use of quantitative gated single photon emission computed tomography (SPECT) as a reference standard.
METHODS Study Patients The study population consisted of 47 consecutive patients (35 men and 12 women; age 29 to 79 years, mean 64 ⫾ 10) who were scheduled for quantitative gated SPECT (QGS) with in sinus rhythm and with suggested IHD. For each patient, 2-dimensional echocardiography and QGS were performed within 48 hours. There were no changes in any patient’s clinical status during this period. Informed written consent was obtained from all patients. The Principles of the AACT Method The AACT method is now commercially available in the newest ultrasound system (SSA-770A Aplio, Toshiba Corp, Tochigi, Japan). The technologic framework of the AACT method has 3 features: (1) weighted separability for edge detection of the endocardial border; (2) contour smoothing using minimization of elastic energy; and (3) partial shape constant contour model. The principles of weighed separability and contour smoothing were described previously.11-15 The newly developed “partial shape constant contour model” consists of 2 features, pattern matching
and contour constraint.16-18 In pattern matching, the intensity pattern of the mitral annulus is stored, and pattern matching against this stored intensity pattern is performed for each frame in the specified measurement range to monitor the position of the mitral annulus (Figure 1, A). This processing is expected to avoid misidentification of the contours because of the chordae tendineae and to improve capability of tracing of the endocardium that can not be clearly displayed. In contour constraint, the endocardial contour extraction range is restricted by 3 sample points, which are placed on both sides of the mitral annulus and the apex (Figure 1, B). This processing is expected to improve trace fitting in cases in which the apical endocardium cannot be clearly visualized. LV Volume and EF Measurement by the AACT Method Echocardiographic measurement of LV volume and EF were performed with an ultrasound system (SSA-770A Aplio, Toshiba Corp) using a 2.5-MHz (2-4 MHz) broadband phased-array transducer. Both apical 4- and 2-chamber views were obtained by means of 2-dimensional echocardiography using harmonic imaging. Clips of echocardiographic images of the LV in these views were acquired and stored digitally in the ultrasound system for offline analysis of LV volume and EF using the AACT method (37 frames/s). In each case, the operator was required to place 3 sample points on both sides of the mitral annulus and the LV apex, in apical views of the end-diastolic image selected with the R wave of the electrocardiogram. Then,
Journal of the American Society of Echocardiography Volume 16 Number 12
Sugioka et al 1239
Figure 2 Steps of automated contour tracking procedure in 4-chamber views: (1) operator marks 3 points at both sides of mitral annulus and at apex in apical views of end-diastolic image that selected with R wave of electrocardiogram; (2) contour of left ventricular (LV) endocardial border is automatically tracked; and (3) extraction of endocardial border of LV cavity is completed in every frame throughout 1 cardiac cycle.
Figure 3 Example of end-diastolic volume, end-systolic volume, and ejection fraction from apical 4- and 2-chamber views by advanced automated contour tracking method. LV, Left ventricular. the endocardial border was identified automatically, and extraction of the endocardial border of the LV cavity was completed in every frame throughout 1 cardiac cycle (Figure 2). The end-systolic frame was selected automatically by estimating smallest cavity volume. In this method, echocardiographic LV volume determined from 4- and 2-chamber apical views and biplane EF were calculated automatically, using disk summation method with a computer incorporated in this ultrasound system (Figure 3). All measurements were performed by a single investigator who was blinded to the gated SPECT results. Unsuccessful tracing was visually defined as cases with obvious discrepancies (ie, tracing inside or outside the LV) between the actual and detected endocardial border. A total of 25 patients were randomly selected for assessment of the intraobserver and interobserver variabilities of the AACT method and for measurement of study time required for analysis of LV volume and EF from an apical 4-chamber image in both the AACT method and manual tracing method. Images obtained in these patients were analyzed by 2 independent observers, without
knowledge of patient identity or previous results. In addition, intraobserver analyses were performed by the same observer 2 weeks after the first evaluation. Interobserver variability was calculated as the SD of the differences between the measurements of 2 independent observers and expressed as a percentage of the average value. Intraobserver variability was calculated as the SD of the differences between the first and second determination for a single observer and expressed as a percentage of the average value. QGS QGS was performed as previously described.19,20 All patients underwent thallium-201-gated SPECT after thallium–201 (111 and 74 MBq for patients without and with previous myocardial infarction, respectively) was injected intravenously. Initial images were obtained immediately after the termination of exercise, and delayed images were obtained 4 hours later. For patients with previous myocardial infarction, an additional dose of 37 MBq thallium–201 was injected at rest immediately after the acquisition of
1240 Sugioka et al
Journal of the American Society of Echocardiography December 2003
Figure 4 Scatter plots showing correlation between advanced automated contour tracking (AACT) method and quantitative gated single photon emission computed tomography (QGS) in measurement of end-diastolic volume (left) and mean end-diastolic volume by AACT and QGS (x axis) and difference between AACT and QGS (y axis) (right). Mean difference and 2 SD limits are shown. SEE, Standard error of the estimate.
delayed images, and the reinjection images were obtained 20 minutes later. SPECT was performed with a 2-detector gamma camera (Vertex, ADAC Laboratories, Milpitas, Calif) equipped with low-energy, general purpose collimators, with the detectors set to form a 90-degree angle. A total of 32 equidistant projections were acquired over 180 degrees in a 64 ⫻ 64 matrix from the 45-degree right anterior oblique projection to the 45-degree left posterior oblique projection with an energy window of 70 keV ⫾ 10%. Electrocardiogram-gated images were acquired with 40-second, 6-degree angular steps. At each projection, 8 frames/ cardiac cycle were acquired, with an allowable change in R-R interval of the mean value ⫾ 20%. Transaxial slices of 4.7-mm-pixel thickness were reconstructed using a Butterworth filter (order, 5.0; critical frequency, 0.35 cycle/ pixel) and the filtered backprojection method (ramp filter) on a processing computer (Pegasys, ADAC Laboratories) with an automatic reorientation algorithm for SPECT (AutoSPECT, Cedars-Sinai Medical Center, Los Angeles, Calif).21 No attenuation correction was applied. A completely automated software program that analyzes LV function (Quantitative Gated SPECT, Cedars-Sinai Medical Center) was used to calculate LV volume and global EF.22,23 Rest LV volume and EF were derived from delayed images for patients without reinjection images and from reinjection images for patients with reinjection images. Statistics Linear regression analysis was used to compare QGS and the AACT assessment of LV volumes in end-diastole and end-systole, and EF. Differences in the measurements obtained by each method are expressed as mean ⫾ SD. Analyses of the differences in measurements were performed using the technique of Bland and Altman.24 We used the paired t test to compare the time required for
LVEF calculation by the AACT method with that required by the manual method. P values ⬍ .05 were considered statistically significant.
RESULTS The diagnoses of the 47 patients were as follows: 14 patients with angina pectoris; 16 with myocardial infarction; 9 with both angina pectoris and myocardial infarction; 3 with dilated cardiomyopathy; and 5 healthy patients. In 41 of 47 patients (87%), automated tracing of the endocardial border was achieved adequately by the AACT method. Of these, there were 29 patients with regional wall-motion abnormalities and 12 without them. Comparison of LV Measurements Between QGS and the AACT Method LV end-diastolic volume measured by the AACT method demonstrated excellent correlation with that by QGS at end-diastole (r ⫽ 0.96, y ⫽ 0.85x ⫹2.4, standard error of the estimate ⫽ 4.2 mL). The mean difference in LV end-diastolic volume between AACT and QGS was 10.8 ⫾ 13.0 mL (Figure 4). LV end-systolic volume measured by the AACT method also demonstrated excellent correlation with that from QGS (r ⫽ 0.98, y ⫽ 0.87x ⫺4.0, standard error of the estimate ⫽ 1.8 mL). The mean difference in LV end-systolic volume measurement between AACT and QGS was 6.7 ⫾ 7.4 mL (Figure 5). LVEF measured by the AACT method correlated well with that by QGS (r ⫽ 0.91, y ⫽ 0.97x ⫹2.4, standard error of the estimate ⫽ 3.7%). The mean difference between AACT and QGS was 0.6 ⫾ 5.5% (Figure 6).
Journal of the American Society of Echocardiography Volume 16 Number 12
Sugioka et al 1241
Figure 5 Scatter plots showing correlation between advanced automated contour tracking (AACT) method and quantitative gated single photon emission computed tomography (QGS) in measurement of end-systolic volume (left) and mean end-systolic volume by AACT and QGS (x axis) and difference between AACT and QGS (y axis) (right). Mean difference and 2 SD limits are shown. SEE, Standard error of the estimate.
Figure 6 Scatter plots showing correlation between advanced automated contour tracking (AACT) method and quantitative gated single photon emission computed tomography (QGS) in measurement of ejection fraction (EF) (left) and mean EF by AACT and QGS (x axis) and difference between AACT and QGS (y axis) (right). Mean difference and 2 SD limits are shown. SEE, Standard error of the estimate.
Time Efficiency and Reproducibility The AACT method required 7 ⫾ 1 seconds for LV volume and EF measurement from 1 set of images during 1 cardiac cycle, whereas 37 ⫾ 4 seconds were needed for the manual tracing method. The study time was significantly shorter using the AACT method than that by manual tracing method (P ⬍ .0001). With the AACT method, intraobserver and interobserver variabilities for LVEF measurement were 4.5% and 7.3%, respectively.
DISCUSSION This is the first study of the biplane, automated assessment of LV volume and EF using the AACT method. In this study, we demonstrated that LVEF was accurately quantified using the AACT method
on patients even when regional wall-motion abnormalities existed. Furthermore, the time needed for LVEF measurement was much shorter using the AACT method than with the manual tracing method. Previous Echocardiographic Estimation of LVEF Two-dimensional echocardiography is a noninvasive and widely accepted method for the measurement of LV function in the clinical setting. LVEF measurement by manual tracing method using apical views, which is recommended by the American Society of Echocardiography,7 is more reliable than that by the M-mode method.2,5 However, clinical use of this method is limited because it requires tedious and time-consuming processes. To improve on these limitations of the manual tracing method, several methods for offline automated border detection
1242 Sugioka et al
have been attempted.8,9 Geiser et al10 reported the autonomous boundary detection for epicardium and endocardium. This method was, however, limited to analysis from the parasternal short-axis view and to cases with high-quality images and with normal LV wall motion. For other offline automated techniques, application in clinical use is also hampered by requirement of processing and much processing time.8,9 Previously, the ACT method was developed for offline automated contour tracing of the LV endocardial boundary without tracing of a region of interest.11-15 The advantages of this method were its simple manual procedure and short processing time. Furthermore, results obtained from this method correlate highly with those of the manual method for estimation of LV cavity areas11 and with left ventriculography for measurement of LVEF in clinical studies.12 However, endocardial border detection by the ACT method often failed because of misidentification of the apex or mitral annulus for patients with suboptimal images. Moreover, in previous studies using the ACT method, patients with wall-motion abnormalities were excluded and the usefulness of this method was not established for patients with IHD.11,12 Advantages of the AACT Method The AACT method especially addresses the limitations of the previous ACT method and has some advantages. First, reliable LVEF measurement by the AACT method can be performed with a high rate of success. During development of the AACT method, technical efforts were focused on improving the accuracy of tracing by using partial shape constant contour model.16-18 This study demonstrated that the success rate of the AACT method in LVEF measurement was 87%, which is high enough for use in the clinical setting. In addition to technologic advancement in the AACT method, the use of harmonic imaging, which allows improvement of LV endocardial border delineation,25 may also contribute to accurate LVEF assessment. Second, the AACT method permits accurate measurement of LVEF even for patients with regional wall-motion abnormalities, because it uses 2 orthogonal planes for biplane LVEF measurement. To our knowledge, no automated method other than the AACT method can measure LVEF in a biplane manner. As presented in this study, biplane LVEF measurements using the AACT method agreed well with those using QGS as the gold standard in patients even when regional wall-motion abnormalities existed. The acoustic quantification method is a widespread method for automated border detection with online analysis.26-28 Although it is a powerful technique that provides real-time LV function, it is applied only in single plane for the assessment of
Journal of the American Society of Echocardiography December 2003
global LV systolic function. In cases with regional LV wall-motion abnormalities, however, LVEF derived from single-plane acoustic quantification may be less accurate compared with that from biplane method.28 By contrast, this AACT method can be applied even to patients with regional wall-motion abnormalities because LV volume is calculated by biplane method in this new automated system. Thus, extended application for patients with IHD for global LV systolic function may be one of the advantages of this method compared acoustic quantification method. Finally, the time of analysis in the AACT method is much shorter than that in the previous ACT method because of an improved calculation algorithm and by recent advancement in computer technology. In this study, study time required for measuring LVEF by the AACT method from 1 apical view was one-fifth that by the manual method. This quick LVEF measurement by the AACT method would be an advantage of this method in the clinical setting. Study Limitations There were a few patients with LVEF less than 35% included in this study. Our data may need to be confirmed in larger patient populations. In this study, the detected boundaries were not compared with any independent technique of boundary detection other than subjective visual interpretation. The accuracy of boundary detection by this AACT method may need to be evaluated in further studies. In addition, in the AACT method, automated detection of endocardial borders can be performed not only in end-systole and end-diastole frames, but in every frame throughout 1 cardiac cycle. In this study, however, we showed only the results of endocardial tracing in end-diastolic and end-systolic frames. Regarding theses frames, we showed a good relationship between AACT and QGS for LV volume. Because the results of endocardial tracing by AACT of endocardial tracing in every frame through a whole cardiac cycle was not shown in this study, it has not been clarified how wall-motion abnormalities affect automated tracing of LV endocardium by the AACT method. Regarding this issue, further investigations are necessary. Finally, although the success rate of the AACT method in LVEF measurement was high, automated tracing of the endocardial border was unable to be achieved adequately in 6 patients for the following reasons: tracing outside the LV lateral wall in 4-chamber view or anterior wall in 2-chamber view at end-diastole in 3 patients; tracing inside the LV in 3 patients; and inadequate tracing in the apical segment in 1 patient. Conclusions The biplane AACT method provides accurate and quick measurement of LVEF for patients even when
Journal of the American Society of Echocardiography Volume 16 Number 12
regional wall-motion abnormalities exist. This newly developed method is reliable in noninvasive evaluation of global LV systolic function. We acknowledge the assistance of Naoki Yoneyama, BS, Hiroyuki Tsujino, MS, Ryoichi Kanda, MS, and Masahide Nishiura, MS, of Toshiba Corporation in technical advices for preparing this manuscript. REFERENCES 1. The Multicenter Postinfarction Research Group. Risk stratification and survival after myocardial infarction. N Engl J Med 1983;309:331-6. 2. Schiller NB, Acquatella H, Ports TA, Drew D, Goerke J, Ringertz H, et al. Left ventricular volume from paired biplane two-dimensional echocardiography. Circulation 1979;60:54755. 3. Folland ED, Parisi AF, Moynihan PF, Jones DR, Feldman CL, Tow DE. Assessment of left ventricular ejection fraction and volumes by real-time, two-dimensional echocardiography. Circulation 1979;60:760-6. 4. Carr KW, Engler RL, Forsythe JR, Johnson AD, Gosink B. Measurement of left ventricular ejection fraction by mechanical cross-sectional echocardiography. Circulation 1979;59: 1196-206. 5. Starling MR, Crawford MH, Sorensen SG, Levi B, Richards KL, O’Rourke RA. Comparative accuracy of apical biplane cross-sectional echocardiography and gated equilibrium radionuclide angiography for estimating left ventricular size and performance. Circulation 1981;63:1075-84. 6. Silverman NH, Ports TA, Snider AR, Schiller NB, Carlsson E, Heilbron DC. Determination of left ventricular volume in children: echocardiographic and angiographic comparisons. Circulation 1980;62:548-57. 7. Schiller NB, Shah PM, Crawford M, DeMaria A, Devereux R, Feigenbaum H, et al. Recommendations for quantitation of the left ventricle by two-dimensional echocardiography. J Am Soc Echocardiogr 1989;2:358-67. 8. Bosch JG, Savalle LH, Burken G, Reiber JH. Evaluation of a semiautomatic contour detection approach in sequences of short-axis two-dimensional echocardiographic images. J Am Soc Echocardiogr 1995;8:810-21. 9. Cootes TF, Taylor CJ, Cooper DH. Active shape models: their training and application. Comput Vis Image Underst 1995; 61:38-59. 10. Geiser EA, Wilson DC, Wang DX, Conetta DA, Murphy JD, Hutson AD. Autonomous epicardial and endocardial boundary detection in echocardiographic short-axis images. J Am Soc Echocardiogr 1998;11:338-48. 11. Hozumi T, Yoshida K, Yoshioka H, Yagi T, Akasaka T, Takagi T, et al. Echocardiographic estimation of left ventricular cavity area with a newly developed automated contour tracking method. J Am Soc Echocardiogr 1997;10:822-9. 12. Sugioka K, Hozumi T, Yagi T, Yamamuro A, Akasaka T, Takeuchi K, et al. Automated quantification of left ventricular function by the automated contour tracking method. Echocardiography 2003;20:313-8.
Sugioka et al 1243
13. Cipolla R, Blake A. The dynamic analysis of apparent contours. Proc 3 Int J Comput Vis 1990;i:616-23. 14. Menet S, Saint-Marc P, Medioni G. Active contour models: overview, implementation and applications. IEEE Proc Int Conf Syst Man Cybern 1990;6:194-9. 15. Kass M, Witkin A, Terzopoulos D. Snakes: active contour models. Int J Comput Vis 1988;1:321-31. 16. Nishiura M, Yuasa M, Watanabe M. Active contour extraction method using partial shape constraint contour model. Syst Comput Jpn 2000;31:20-8. 17. Nishiura M, Furukawa R, Tsujino H, Kanda R. Automated tracking of the mitral annulus using two-dimensional pattern matching technique: a validation study [abstract]. J Am Soc Echocardiogr 2001;14:469. 18. Tsujino H, Nishiura M, Furukawa R, Kanda R. Quantitative assessment of the mitral annular motion using a two-dimensional automated annular tracking method [abstract]. J Am Soc Echocardiogr 2001;14:478. 19. Yamagishi H, Shirai N, Yoshiyama M, Teragaki M, Akioka K, Takeuchi K, et al. Incremental value of left ventricular ejection fraction for detection of multivessel coronary artery disease in exercise 201-Tl gated myocardial perfusion imaging. J Nucl Med 2002;43:131-9. 20. Shirai N, Yamagishi H, Yoshiyama M, Teragaki M, Akioka K, Takeuchi K, et al. Incremental value of assessment of regional wall motion for detection of multivessel coronary artery disease in exercise 201-Tl gated myocardial perfusion imaging. J Nucl Med 2002;43:443-50. 21. Germano G, Kavanagh PB, Su H, Mazzanti M, Kiat H, Hachamovitch R, et al. Automatic reorientation of threedimensional, transaxial myocardial perfusion SPECT images. J Nucl Med 1995;36:1107-14. 22. Germano G, Kiat H, Kavanagh PB, Moriel M, Mazzanti M, Su HT, et al. Automated quantification of ejection fraction from gated myocardial perfusion SPECT. J Nucl Med 1995; 36:2138-47. 23. Germano G, Erel J, Lewin H, Kavanagh PB, Berman DS. Automated quantitation of regional myocardial wall motion and thickening from gated technetium-99m sestamibi myocardial perfusion single-photon emission computed tomography. J Am Coll Cardiol 1997;30:1360-7. 24. Bland JM, Altman DG. Statistical methods for assessing agreement between two methods of clinical measurement. Lancet 1986;1:307-10. 25. Spencer KT, Bednarz J, Rafter PG, Korcarz C, Lang RM. Use of harmonic imaging without echocardiographic contrast to improve two-dimensional image quality. Am J Cardiol 1998; 82:794-9. 26. Vandenberg BF, Rath LS, Stuhlmuller P, Melton HE Jr, Skorton DJ. Estimation of left ventricular cavity area with an on-line, semiautomated echocardiographic edge detection system. Circulation 1992;86:159-66. 27. Perez JE, Waggoner AD, Barzilai B, Melton HE Jr, Miller JG, Sobel BE. On-line assessment of ventricular function by automatic boundary detection and ultrasonic backscatter imaging. J Am Coll Cardiol 1992;19:313-20. 28. Bednarz JE, Marcus RH, Lang RM. Technical guideline for performing automated border detection studies. J Am Soc Echocardiogr 1995;8:293-305.