Full terms and conditions of use:

Transcription

1 This article was downloaded by:[uppsala Universitetsbibiotek] [Uppsala Universitetsbibiotek] On: 3 May 2007 Access Details: [subscription number ] Publisher: Informa Healthcare Informa Ltd Registered in England and Wales Registered Number: Registered office: Mortimer House, Mortimer Street, London W1T 3JH, UK Acta Radiologica Publication details, including instructions for authors and subscription information: The Exactness of Left Ventricular Segmentation in Cine Magnetic Resonance Imaging and Its Impact on Systolic Function Values To cite this Article:, 'The Exactness of Left Ventricular Segmentation in Cine Magnetic Resonance Imaging and Its Impact on Systolic Function Values', Acta Radiologica, 48:3, To link to this article: DOI: / URL: PLEASE SCROLL DOWN FOR ARTICLE Full terms and conditions of use: This article maybe used for research, teaching and private study purposes. Any substantial or systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expressly forbidden. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. Taylor and Francis 2007

2 ORIGINAL ARTICLE The Exactness of Left Ventricular Segmentation in Cine Magnetic Resonance Imaging and Its Impact on Systolic Function Values ACTA RADIOLOGICA C. EBELING BARBIER, L. JOHANSSON, L. LIND, H. AHLSTRÖM & T. BJERNER Department of Radiology and Department of Medicine, Uppsala University Hospital, Uppsala, Sweden; AstraZeneca, Gothenburg, Sweden Barbier CE, Johansson L, Lind L, Ahlström H, Bjerner T. The exactness of left ventricular segmentation in cine magnetic resonance imaging and its impact on systolic function values. Acta Radiol 2007;48: Purpose: To evaluate the impact of exactness of the segmentation of the left ventricle (LV), using cine magnetic resonance imaging (MRI). Material and methods: Steady-state free-precession cine MRI was performed on 100 randomly selected subjects. Myocardial borders were outlined on short-axis images using three methods: method 1 was computer assisted, excluding papillary muscles from the left ventricular mass (LVM); method 2 was similar but included papillary muscles; and method 3 was manually traced including papillary muscles. LV end-systolic (ES) and end-diastolic (ED) masses and volumes, ejection fraction (EF), stroke volume (SV), and cardiac output (CO) were calculated from these measurements. The difference between the ES and ED LVM was used to estimate the exactness of the methods. Results: Method 3 was the most exact, and method 1 was the least exact. The three methods generated differing EF, SV, and CO measurements. With an ES ED LVM difference exceeding 20 g, the mean SV measurement error was ml. Conclusion: Manual tracing proved more exact than computer-assisted quantification. Exactness had an impact on EF, SV, and CO measurements, and the ES ED LVM difference can be used to identify assessments that would benefit from more exact segmentation. Key words: Adults; cardiac; heart; left ventricle; MR imaging; segmentation Charlotte Ebeling Barbier, Department of Radiology, Uppsala University Hospital, SE Uppsala, Sweden (tel , fax , . charlotte.ebeling_barbier@radiol.uu.se) Accepted for publication 8 December 2006 Accurate assessment of myocardial mass and ventricular volume is an important diagnostic and prognostic tool in patients suffering from various forms of heart disease. Different modalities with accompanying limitations have been used. Cardiac magnetic resonance imaging (MRI), with its high spatial resolution, accurately determines left ventricular mass (LVM) when compared to postmortem mass in animal models (5, 7, 10, 19). The excellent blood/myocardial contrast of steady-state free-precession sequences enables better endocardial border definition and lower interobserver variability of the assessments compared to turbo gradient-echo sequences (3, 12). Commercially available software calculates LVM, LV volumes, and systolic function values based on outlining of the myocardial borders. This is commonly performed in contiguous short-axis cine images in diastole and systole (10). The end-diastolic frame is generally used for mass determination based on the assumption that the partial volume effects should be less of a problem (10). Outlining is accomplished either with manual tracing using a mouse or with varying semiautomatic or automatic segmentation techniques. The former enables a more accurate assessment, whereas the latter are more prompt but rather coarse, and thus often require manual adjustment (12). Manual tracing is accordingly proved to estimate LVM more accurately than autosegmentation without adjustment in dogs and pigs, but requires six times the analysis time (6). Manually corrected autosegmentation determines LVM within 3% of the postmortem value in dogs (5). Some include papillary muscles in the LVM (1, 5, 6, 10, 16) while others include papillary muscles in DOI / # 2007 Taylor & Francis

3 286 C. Ebeling Barbier et al. the LV blood volume (2, 4, 14, 19), and there is no consensus on the significance of crypts and trabeculae, which demand manual tracing. Depending on whether papillary muscles and trabeculae are included in the LVM or not, differences are found in stroke volume (SV), end-systolic volume (ESV), and left ventricular end-diastolic volume (EDV), but not in ejection fraction (EF) or right ventricular EDV (15). However, outlining of papillary muscles and trabeculae almost doubles the manual analysis time (15). Exactness is hence achieved at the expense of time, which raises the issue regarding what impact the exactness of the segmentation has. The aim of this study was to evaluate the exactness of three different methods for quantification of LVM and LV function. A second aim was to assess the impact of the exactness of the segmentation on the obtained systolic function values. Material and Methods Study population Cardiac MRI was performed, after approval from the local ethics committee and written informed consent, on an unselected population sample (52 women, 48 men) from the Uppsala region of Sweden who were recruited at 70 years of age in the Prospective Investigation of the Vasculature in Uppsala Seniors (PIVUS) study (9). In this population, 10% suffered from angina pectoris, 7% had had a myocardial infarction, 5% had been submitted to interventional treatment of the coronary arteries, 4% had had a stroke, 4% suffered from congestive heart failure, and 35% were on antihypertensive treatment. The subjects included in the present study constitute a subpopulation composed of the first 100 consecutively analyzed subjects of the 250 who were planned to undergo cardiac MRI. Image acquisition Imaging was performed on a 1.5T MRI system (Gyroscan Intera; Philips Medical Systems, Best, The Netherlands) with a 25 mt/m gradient system, using the standard SENSE cardiac coil in the supine position and retrospectively gated vector ECG for cardiac triggering. After initial surveys, acquisitions were obtained using a steady-state free-precession cine sequence covering the left ventricular myocardium in 8-mmthick short-axis slices with 2.5 mm slice gap. The number of slices was adjusted to cover the heart from the apex to the atria. Two slices were acquired per breath-hold (14 s) with an acquired in-plane resolution of mm (reconstructed to mm). The following parameters were used: TR5shortest (,3.6 ms), TE5shortest (,1.8 ms), flip angle 70, bandwidth Hz/pixel, 18 phases/cardiac cycle, field of view (FOV) 400 mm, matrix 256, parallel imaging (SENSE) factor 2, and k-lines segments (TFE factor) 19. Image analysis Image analysis was performed on a workstation using commercially available analysis software (ViewForum; Philips Medical Systems, Best, The Netherlands). Quantification was made on shortaxis images that were zoomed to facilitate detailed segmentation. The computer performed interpolation in order to bridge the slice gaps. The first phase following the ECG R-wave was defined as end diastole (ED), and end systole (ES) was defined visually as the phase with the smallest left ventricular volume. The same observer performed all outlining. In method 1 (Fig. 1A), border definition of the epicardial contour was accomplished by manual tracing using a mouse or a pen tablet (Wacom, Saitama, Japan). The endocardial contour was generated with computer assistance, where a manually drawn contour is automatically adapted to the underlying image. The papillary muscles were included in the endocardial contour when attached to the ventricular wall. Papillary muscles not attached to the wall were included in the LV blood volume. Systolic function values (i.e., ejection fraction (EF), stroke volume (SV), and cardiac output (CO)) and the ED and ES LVMs were calculated by the software using Simpson s rule approximation and assuming a myocardial density of 1.05 g/ml. In method 2 (Fig. 1B), the same endocardial and epicardial borders were used. Papillary muscles were manually outlined separately when not attached to the ventricular wall, and all papillary muscles were included in the LVM as the software computed the systolic function values, the LVM, and the LV volumes. The same contour set was hence used for measurements in methods 1 and 2, the only difference being the exclusion or inclusion of the separately outlined papillary muscles in the LVM. In method 3 (Fig. 1C), the same, manually established, epicardial contour was used, whereas the endocardial and the papillary muscle contours were manually corrected in order to include crypts and trabeculae to the greatest possible extent. A second contour set was hence established. LVM and systolic function values were computed.

4 Exactness of Left Ventricular Segmentation Using Cine MRI 287 A B C Fig. 1. A. Short-axis cine magnetic resonance images in diastole illustrating the computer-assisted contour set used in method 1, where only the papillary muscles included in the endocardial contour were included in the left ventricular mass (LVM). B. The same contour set was used in method 2, but with all papillary muscles included in the LVM. C. In method 3, all contours were manually traced, and all papillary muscles and the endocardial trabeculae were included in the LVM. In all three methods, slices at the base of the heart where ventricular myocardium was present were considered to be within the left ventricle, sometimes resulting in the epi- and endocardial contours being superimposed along the part of the border that was not surrounded by myocardium (Fig. 2). If the aortic valve appeared in the basal slice, the contours were drawn up to the valve and joined by a straight line through the blood pool (1) (Fig. 3A). However, if the subaortic blood volume appeared tubular and was not surrounded by myocardium, the contours were joined by a straight line at the junction of the LV cavity and the tubular subaortic area (Fig. 3B). This approach was chosen to give the most exact estimation of the blood volume. To be able to draw a straight line, the computer assistance tool had to be switched off so that all contours in the first subaortic slice or slices were manually traced in all three methods. The difference between ES and ED LVM was calculated by subtracting the ED LVM from the ES LVM. This was used to assess the exactness of the three methods, based on the assumption that the ES ED LVM difference should be as small as possible (16, 17). The method is hence considered more exact the lesser the difference. All measurements except EF were indexed to body surface area (BSA). Subjects were grouped based on ES ED LVM difference in the measurements generated with method 1 (before indexation to BSA). Those with a difference smaller than 20 g were put into one group, and those with a difference larger than 20 g were put into the other. These two groups were compared regarding the difference in SV between methods 1 and 3. Analysis time was recorded at five random occasions of establishing the two contour sets. Fig. 2. Short-axis cine magnetic resonance image in systole illustrating the epi- and endocardial contours being superimposed along the part of the border that is not surrounded by myocardium. Statistical analysis Mean values standard deviations (SD) were calculated for all parameters using StatView (SAS Institute, Cary, N.C., USA). Variance analysis of differences between the three methods was performed using ANOVA for repeated measurements and Fisher s PLSD for post hoc analysis. The unpaired t-test was used to analyze the stroke volume difference between the two groups with ES ED LVM difference larger or smaller than 20 g. The significance level was set at 0.05 for all analyses.

5 288 C. Ebeling Barbier et al. A Fig. 3. A. Short-axis cine magnetic resonance images in systole illustrating how contours were drawn up to the aortic valve and joined by a straight line through the blood pool in cases where the aortic valve appeared in the basal slice. B. In cases where the subaortic blood volume appeared tubular and was not surrounded by myocardium, the contours were joined by a straight line at the junction of the LV cavity and the tubular subaortic area. Results Results are presented as mean values SD before and after indexation to body surface area (BSA) in Tables 1 and 2. B The ES ED LVM difference was greatest in the measurements obtained using method 1, less in the measurements obtained using method 2, and least in the measurements obtained using method 3, before and after indexation to BSA. This variation Table 1. Mean values standard deviation for end-diastolic and end-systolic left ventricular masses, the difference between these, and the systolic function values obtained using three different methods of quantification Method ED LVM, g ES LVM, g Difference ES ED LVM, g EF, % SV, ml CO, l/m Difference between methods 1 vs. 2 ns ns P ns ns ns 1 vs. 3 ns ns Pv P ns ns 2 vs. 3 ns ns P P ns ns ED: end diastolic; ES: end systolic; LVM: left ventricular mass; EF: ejection fraction; SV: stroke volume; CO: cardiac output; ns: nonsignificant. Method 1: computer-assisted outlining excluding papillary muscles from the LVM; method 2: similar to method 1 but including papillary muscles; method 3: manually outlined including papillary muscles. Table 2. Mean values standard deviation indexed to body surface area (BSA) for end-diastolic and end-systolic left ventricular masses, the difference between these, stroke volume, and cardiac output obtained using three different methods of quantification Method ED LVM, g /BSA ES LVM, g /BSA Difference ES ED LVM, g /BSA SV, ml /BSA CO, l/m /BSA Difference between methods 1 vs. 2 ns ns P ns ns 1 vs. 3 ns ns Pv P P vs. 3 ns ns P ns ns ED: end diastolic; ES: end systolic; LVM: left ventricular mass; SV: stroke volume; CO: cardiac output; ns: non-significant. Method 1: computer-assisted outlining excluding papillary muscles from the LVM; method 2: similar to method 1 but including papillary muscles; method 3: manually outlined including papillary muscles.

6 Exactness of Left Ventricular Segmentation Using Cine MRI 289 in ES ED LVM difference was significant between all three methods (Tables 1 and 2). When splitting for gender, the ES ED LVM differences remained significant in women between methods 3 and 1 and between methods 3 and 2, but not between methods 2 and 1. However, after indexation to BSA, differences were significant between all methods. In men, the ES ED LVM differences were significant between methods 3 and 1 and between methods 2 and 1, but not between methods 3 and 2. This remained after indexation to BSA. The group with an ES ED LVM difference smaller than 20 g (70 subjects) presented a mean difference in SV of ml, whereas the group with an ES ED LVM difference larger than 20 g (30 subjects) presented a mean difference in SV of ml. The groups were significantly different (Pv0.0001) regarding SV difference. These results are illustrated in Fig. 4. The mean analysis time for the first contour set, generating measurements for methods 1 and 2, was 13 min. The second contour set, generating method 3, added on average 10 min. Discussion The impact of the exactness of the segmentation of the left ventricle is a current matter of discussion, and there is a demand for uniform criteria for outlining the endocardial contour (11). It has been argued that inclusion of papillary muscles and trabecular volumes in LVM generates differences Fig. 4. Graph of the subjects grouped according to the difference in end-systolic (ES) and end-diastolic (ED) left ventricular mass in the values generated with computer assistance excluding the papillary muscles (method 1), revealing a significant difference in stroke volume (SV) between the groups (Pv0.0001). that are too small to be clinically relevant (15). Others have found significant differences in EF, LVM, and volumes but still recommend a tracing convention excluding the trabeculae because of its superior reproducibility (11). The exactness of the segmentation may also be influenced by acquisition factors such as the number of phases recorded per cardiac cycle. It has been reported that at least 11 phases are needed to maintain accuracy of measurements of LV volumes, EF, and mass (13). In the present study, 18 phases were recorded, which allows two slices per breathhold and is the standard amount used in clinical routine at our clinic. In the present study, method 3 was concluded to be most exact, method 2 less exact, and method 1 the least exact when using the ES ED LVM difference as a reference. This conclusion is based on experimental data proving the LV myocardium incompressible throughout diastole and systole (17), which is supported by echocardiography (8) and MRI (16) LVM determinations in humans. Since LVM is calculated from myocardial volume measurements, it should be equal in ED and ES. To the best of our knowledge, using the ES ED LVM difference to estimate exactness in this way has not been described before. The partial volume effect sensitivity at the base and apex of the heart and the influence of the angulations of the slices are well-known problems associated with short-axis segmentation. Despite the intramyocardial blood volume being 20% larger in diastole than in systole (18), MRI determines the ES LVM to be greater than the ED LVM (16). This has been attributed to partial volume effects (10), which have a larger impact in ES than in ED, as proved by the observation that the ES LVM is considerably larger when assessed only in the short axis than in both the short and long axes, whereas there is no difference in the ED LVM (16). Using short-axis segmentation, ES LVM is hence generally overestimated due to partial volume effects. The ES LVM was also determined greater than the ED LVM in the present study: on average 14.6% using the least exact method, 11.5% using the more exact one, and 8.8% using the most exact one. These results are consistent with others who have determined the ES LVM to be 14.7% greater than the ED LVM when manually tracing in the short axis (16). Recommendations to add segmentation in the long axis to reduce the partial volume effects have been made (3, 16). It is also possible to force agreement between ES LVM and ED LVM by including or excluding slices at the base and apex of the heart, a method which is commonly used (16).

7 290 C. Ebeling Barbier et al. In the present study, however, these partial volume effects should have equal impact on the three methods, since the outer, apical, and basal boundaries were the same. The variation between the methods in ES ED LVM difference can hence not be entirely explained by partial volume effects, and the conclusion that it reflects variations in exactness of the segmentation of the left ventricle seems appropriate. The present study reports, consistent with PAPAVASSILIU et al. (11), that the exactness of the segmentation has an impact on the measurements of systolic function. The differences in these measurements are, however, small enough to be within the range of normal variation, and the least exact method would be sufficient for routine determination of systolic function (EF, SV, and CO). However, in cases where the least exact method generated large ES ED LVM differences, i.e., the exactness was poor, manual tracing could entail a difference in stroke volume larger than 10 ml, and the most exact method could be necessary to avoid misinterpretations. This calls for a cut-off value to identify those who would benefit from manual tracing. In order to find subjects with a possible error in SV of 10 ml or more, the cut-off value was set to a 20-g ES ED LVM difference. The measurements not indexed to BSA were used, since these are the ones presented by the commercially available software and hence used in clinical practice. With one exception, all subjects with a difference smaller than 20 g displayed a SV difference, i.e., possible error, smaller than 10 ml (mean 4.8 ml; Fig. 4). This 20-g cut-off value can be useful in clinical practice, applying the following approach. Start with computer-assisted segmentation, excluding the papillary muscles, then calculate the ES ED LVM difference to estimate the exactness. If this difference is larger than 20 g, inspect the basal and apical slices of the ES and ED frames and force agreement by including or excluding slices if they do not correspond. If the ES ED LVM difference still remains larger than 20 g, manual tracing can be performed in order to avoid an error in SV large enough to be able to cause misinterpretations. Apart from the above-discussed problems associated with the short-axis approach, the present study was limited by the fact that only semiautomatic and manual techniques were compared. However, fully automatic segmentation techniques frequently require manual adjustment (12), and consequently varying forms of semiautomatic techniques are advocated (5, 19). Furthermore, the computer assistance offered by the used software to establish the epicardial contour is inadequate (12), and computer assistance was not optional for the papillary muscle contour. These contours were hence manually traced, mainly leaving the endocardial and, to some extent, the papillary contours to differ between methods 2 and 3. Another limitation was that reproducibility could not be evaluated since measurements were not repeated and were performed only by one observer. In conclusion, the results from the present study suggest that the exactness of the segmentation has an impact on the systolic function values. The exactness can be adapted to fit varying clinical demands, and the ES ED LVM difference can be used to identify the assessments that would benefit most from more exact segmentation. Acknowledgments This study was supported by the Swedish Research Council, grant no. K X A, and the Linné Foundation for Medical Research, Uppsala, Sweden. References 1. Alfakih K, Plein S, Thiele H, Jones T, Ridgway JP, Sivananthan MU. Normal human left and right ventricular dimensions for MRI as assessed by turbo gradient echo and steady-state free precession imaging sequences. J Magn Reson Imaging 2003;17: Alfakih K, Reid S, Jones T, Sivananthan M. Assessment of ventricular function and mass by cardiac magneticresonance imaging. Eur Radiol 2004;14: Bloomer TN, Plein S, Radjenovic A, Higgins DM, Jones TR, Ridgway JP, et al. Cine MRI using steady state free precession in the radial long axis orientation is a fast accurate method for obtaining volumetric data of the left ventricle. J Magn Reson Imaging 2001;14: Dornier C, Somsen GA, Ivancevic MK, Osman NF, Didier D, Righetti A, et al. Comparison between tagged MRI and standard cine MRI for evaluation of left ventricular ejection fraction. Eur Radiol 2004;14: Fieno DS, Jaffe WC, Simonetti OP, Judd RM, Finn JP. TrueFISP: assessment of accuracy for measurement of left ventricular mass in an animal model. J Magn Reson Imaging 2002;15: Francois CJ, Fieno DS, Shors SM, Finn JP. Left ventricular mass: manual and automatic segmentation of true FISP and FLASH cine MR images in dogs and pigs. Radiology 2004;230: Furber A, Balzer P, Cavaro-Menard C, Croue A, Da Costa E, Lethimonnier F, et al. Experimental validation of an automated edge-detection method for a simultaneous determination of the endocardial and epicardial borders in short-axis cardiac MR images: application in normal volunteers. J Magn Reson Imaging 1998;8:

8 Exactness of Left Ventricular Segmentation Using Cine MRI King DL, Coffin Lel K, Maurer MS. Noncompressibility of myocardium during systole with freehand three-dimensional echocardiography. J Am Soc Echocardiogr 2002;15: Lind L, Fors N, Hall J, Marttala K, Stenborg A. A comparison of three different methods to evaluate endothelium-dependent vasodilation in the elderly: the Prospective Investigation of the Vasculature in Uppsala Seniors (PIVUS) study. Arterioscler Thromb Vasc Biol 2005;25: Lorenz CH, Walker ES, Morgan VL, Klein SS, Graham TP Jr. Normal human right and left ventricular mass, systolic function, and gender differences by cine magnetic resonance imaging. J Cardiovasc Magn Reson 1999;1: Papavassiliu T, Kuhl HP, Schroder M, Suselbeck T, Bondarenko O, Bohm CK, et al. Effect of endocardial trabeculae on left ventricular measurements and measurement reproducibility at cardiovascular MR imaging. Radiology 2005;236: Plein S, Bloomer TN, Ridgway JP, Jones TR, Bainbridge GJ, Sivananthan MU. Steady-state free precession magnetic resonance imaging of the heart: comparison with segmented k-space gradient-echo imaging. J Magn Reson Imaging 2001;14: Roussakis A, Baras P, Seimenis I, Andreou J, Danias PG. Relationship of number of phases per cardiac cycle and accuracy of measurement of left ventricular volumes, ejection fraction, and mass. J Cardiovasc Magn Reson 2004;6: Sandstede J, Lipke C, Beer M, Hofmann S, Pabst T, Kenn W, et al. Age- and gender-specific differences in left and right ventricular cardiac function and mass determined by cine magnetic resonance imaging. Eur Radiol 2000;10: Sievers B, Kirchberg S, Bakan A, Franken U, Trappe HJ. Impact of papillary muscles in ventricular volume and ejection fraction assessment by cardiovascular magnetic resonance. J Cardiovasc Magn Reson 2004;6: Swingen C, Wang X, Jerosch-Herold M. Evaluation of myocardial volume heterogeneity during end-diastole and end-systole using cine MRI. J Cardiovasc Magn Reson 2004;6: Tsuiki K, Ritman EL. Direct evidence that left ventricular myocardium is incompressible throughout systole and diastole. Tohoku J Exp Med 1980;132: Wu EX, Tang H, Wong KK, Wang J. Mapping cyclic change of regional myocardial blood volume using steady-state susceptibility effect of iron oxide nanoparticles. J Magn Reson Imaging 2004;19: Young AA, Cowan BR, Thrupp SF, Hedley WJ, Dell Italia LJ. Left ventricular mass and volume: fast calculation with guide-point modeling on MR images. Radiology 2000;216: