International Journal of Cardiology 124 (2008) 47 – 56 www.elsevier.com/locate/ijcard
Objective quantification of global and regional left ventricular systolic function by endocardial tracking of contrast echocardiographic sequences ☆ Javier Bermejo a,⁎, Jonathan Timperley b , Rodolfo G. Odreman a,1 , Miguel Mulet c , J. Alison Noble d , Adrian P. Banning b , Raquel Yotti a , Esther Pérez-David a , Jérôme Declerck c , Harald Becher b , Miguel A. García-Fernández a a
Department of Cardiology, Hospital General Universitario Gregorio Marañón, Madrid, Spain b Department of Cardiology, The John Radcliffe Hospital, Oxford, United Kingdom c Mirada Solutions Ltd, Oxford, United Kingdom d Department of Engineering Science, University of Oxford, United Kingdom
Received 27 March 2006; received in revised form 29 August 2006; accepted 30 December 2006 Available online 20 April 2007
Abstract Background: Echocardiographic assessment of LV wall motion is still most frequently done visually. This study was designed to validate a new system for semi-automatic quantification of global and regional LV systolic function from contrast-enhanced cross-sectional echocardiograms. Methods: Measurements of LV volumes were validated in 50 patients using magnetic resonance (MR) as reference. The regional identification of the endocardial boundary was validated frame-by-frame against the visually identified border in another 27 patients. Finally, the applicability of the system for quantifying stress-echocardiographic exams was assessed in 52 patients undergoing dobutamine interventions. Echocardiographic sequences were digitally processed using custom-built algorithms, based on local phase feature descriptors, deformable contour fitting, and prospective training. Results: Compared to MR, the tracing system showed reasonable accuracy, with relative errors for end-diastolic volume, end-systolic volume, and EF of 21 ± 20%, 27 ± 33%, and − 4 ± 18%, respectively. Regional agreement of the instantaneous contours with visually traced borders was within the limits of visual reproducibility. The system was suitable for tracking stress-echo studies from all patients except two (96%). Quantification of regional radial shortening allowed to discriminate segments showing an abnormal regional wall motion with an overall area under the ROC curve of 0.87. Conclusions: A reliable and accurate quantification of LV systolic function can be obtained by processing contrast echocardiograms. Values of LV volumes, ejection fraction, and regional endocardial shortening adequately correlate with currently available reference methods. Readily applicable to baseline and stress studies, endocardial tracking techniques increase the reliability of echocardiography for the assessment of global and regional systolic function. © 2007 Elsevier Ireland Ltd. All rights reserved. Keywords: Echocardiography; Systolic function; Imaging; Digital processing; Left ventricle
☆
Supported in part by the European Community, Information Societies Technology Programme of the Commission of the European Communities, Contract Number IST-1999-10837. ⁎ Corresponding author. Cardiología No Invasiva, Servicio de Cardiología, Hospital General Universitario Gregorio Marañón, Dr. Esquerdo 46, 28007 Madrid, Spain. Tel.: +34 91 5868279; fax: +34 91 5866727. E-mail address:
[email protected] (J. Bermejo). 1 Current address: Instituto de Investigaciones Cardiovasculares, Universidad de Los Andes, Mérida, Venezuela. 0167-5273/$ - see front matter © 2007 Elsevier Ireland Ltd. All rights reserved. doi:10.1016/j.ijcard.2006.12.091
48
J. Bermejo et al. / International Journal of Cardiology 124 (2008) 47–56
An accurate assessment of global left ventricular (LV) systolic function is essential to guide therapy and determine prognosis in most cardiovascular diseases. Also, the analysis of LV regional wall motion (RWM), either at baseline or during stress, is the basis for the diagnosis of ischemic heart disease by means of echocardiography. The assessment of LV volumes, ejection fraction (EF) and RWM relies on the identification of the endocardial border of the LV cavity. However, poor ultrasonic window precludes optimal border visualization in more than 15% of patients [1]. Current technology of ultrasound scanners and intravenous echocardiographic contrast agents have significantly improved endocardial visualization in the clinical setting [2,3]. However, even with these improvements, quantification of LV volumes still relies on manual tracing of the endocardial boundaries. For the purpose of stress-echocardiography, RWM is also assessed visually, leading to important variability among observers and institutions [4]. Although some attempts have been taken for developing a relatively user-independent quantification of LV volumes from contrast echocardiographic sequences [5–7], no study has yet ascertained the value of a system capable of providing simultaneous measurements of global and regional function from contrast echocardiograms in the clinical setting. Current image processing technology has developed important advances in semiautomatic detection of image features. The present study was designed to test the applicability and accuracy of a new system for semiautomatic quantification of LV volumes and RWM, based on postprocessing contrast echocardiographic two-dimensional sequences. A consecutive step-by-step approach was followed, aiming particular targets in different populations. The specific objectives were: 1) to validate against magnetic resonance (MR) the accuracy of LV volumes and EF provided by the system, and quantify its impact on inter and intraobserver variability (study I), 2) to assess the accuracy of the regional identification of the endocardial boundary against the visually traced endocardial position throughout the full cardiac cycle (study II), and 3) to test the applicability
of the system in consecutive patients undergoing pharmacological stress echocardiography, and compare quantitative values of RWM to the visual analysis of expert readings in this setting (study III). 1. Methods 1.1. Study patient groups Three different population study groups were used for the three major objectives (Table 1). For validation against MR images (study I), 50 patients were recruited (group I). This population was selected at two different institutions, based on different inclusion criteria: 1) clinical indication of a MR examination for a quantitative measurement of LV systolic function (n = 28 patients selected at Center A; group I-A), or 2) clinical indication of a contrast echocardiographic examination (n = 22 patients selected at Center B; group I-B). To assess the accuracy of RWM analysis (study II), 27 consecutive patients with clinical indication of a contrast echocardiographic examination due to poor ultrasonic window were enrolled at Center A. Finally, to test the system against visual analysis of regional function during stress-echo (study III), 52 consecutive patients undergoing pharmacological dobutamine-echo were included at Center A. Exclusion criteria for all patient groups were: 1) an acute myocardial infarction within the preceding 2 weeks, 2) atrial fibrillation, and 3) a known or suspected right-to-left cardiac shunt. Patients carrying an implanted pacemaker or cardioverter–defibrillator, were excluded from group I. Local Research Ethics Committees approved the study protocols and written informed consent was given by all patients. 1.2. Contrast echocardiography image acquisition All exams were performed either with a Sonos 5500 (Philips Medical Systems, Andover MA) or an Acuson Sequoia c-256 ultrasound scanners (Siemens, Germany), equipped with broad-band transducers and low-power
Table 1 Patient study groups and clinical data Group I
Objective Reference technique Enrollment site Type of examination Number of patients Number of echo sequences/segments Age (years; mean ± SD) Male/female RWM abnormalities (n; %)
Group I-A
Group I-B
Global validation of volumes and EF MR Center A Rest 28 56 60 ± 13 21/7 24 (86%)
Global validation of volumes and EF/reproducibility MR Center B Rest 22 44 57 ± 13 21/1 16 (73%)
EF: Ejection fraction; MR: magnetic resonance; RWM: regional wall motion.
Group II
Group III
Regional validation
Quantification of RWM
Visual tracing contrast echo Center A Rest 27 60/360 69 ± 14 18/9 17 (63%)
Visual scoring contrast echo Center A Stress-echo 52 417/2502 68 ± 7 35/15 31 (62%)
J. Bermejo et al. / International Journal of Cardiology 124 (2008) 47–56
49
harmonic-image modalities (mechanical index 0.1 to 0.3). Sulphur Hexafluoride (Sonovue, Bracco, Italy) was used for study I, whereas perflutren protein-Type A microspheres (Optison, General Electric) were used for studies II and III. Contrast was administered either as repetitive bolus (1–2 ml of Sonovue, or 0.5–1 ml of Optison [groups I-A, II and III]), or as a continuous infusion (0.8 ml/min, up to 15 ml [group I-B]). Frequency of bolus and infusion rate were modified by the examiner, targeting for optimal opacification. Threebeat parasternal short axis and apical four-, and twochamber views were used for processing, and scanners were optimized to obtain acquisition frame rates ≥ 25 Hz. For study III, a standard dobutamine–atropine stress-echo protocol was performed [8], and contrast sequences were obtained at baseline, at low-dose (10 μg/kg/min), and at peak dobutamine dose (achievement of target heart rate, or dobutamine at 40 μg/kg/min with or without atropine).
ms (group I-B, Center B). All scans were ECG gated and performed using a five or six element spine array coil combined with a 2 channel flex array placed over the chest (Siemens). Scout images were acquired to correctly identify the true long axis of the left ventricle followed by true fast imaging with steady state free precession procession (TrueFISP) cine images in both 2-chamber and 4-chamber orientation. Diastolic frames from these images were used to identify the atrio-ventricular groove to align the base of the LV in the short axis plane. Parallel slices were then acquired from apex to base to cover the entire ventricle. Image analysis was performed using either Philips Easy Vision or Siemens ARGUS software. For group I-B, measures were performed by 2 operators and repeated by one of them 4 weeks later. Operators were blinded to the echocardiogram.
1.3. MR image acquisition and processing (study I)
Ultrasound image analysis was processed using the Quamus® software developed by Mirada Solutions Ltd (Oxford, UK) in conjunction with the University of Oxford, UK [9,10]. After digital transfer to a dedicated computer, study initialization is performed either using 3 (base and apex, in apical views) or 5 (equally spaced, in short axis views) control points which are matched automatically to the endocardial boundary (Fig. 1). Matching to the endocardial shape is then performed using local phase feature descriptors [11–13]. The features are used as targets for the contour using an adaptation of the iterative closest
The MR examination was performed within 4 h of echocardiography in all patients except in one (performed 24 h later due to technical difficulties). This patient suffered no clinical events between the imaging techniques. MR studies were performed either on a Philips Intera 9.1 1.5 T scanner with a gradient performance of 30 mT/m and a gradient slew rate of 150 mT/m/ms (group I-A, Center A) or a Sonata 1.5 T scanner (Siemens, Germany), with gradient a performance of 40 mT/m and gradient slew rate of 200 mT/m/
1.4. Ultrasound image analysis
Fig. 1. Image-processing algorithm of the tracking system. Panel A: Raw end-diastolic frame as obtained from the ultrasound system. Panel B: Result of initial non-linear image filtering, segmentation for feature enhancement and contour fitting. Panel C: Overlay of the tracked result (end-diastolic frame in green; end-systolic frame in red-orange and blue) on the original image for visual verification. Panel D: Regional analysis of radial fractional shortening for basal, mid and apical septum (BS, MS and AS, respectively), and apical, mid and basal lateral (AL, ML and BL) walls, respectively. Apical akinesis is readily visualized.
50
J. Bermejo et al. / International Journal of Cardiology 124 (2008) 47–56
point technique [11]; this adaptation also includes statistical shape modeling [12] to increase robustness in shadowed areas. A database of manually drawn contours on a variety of patients (30 in each of the views) was used to train the system to follow endocardial motion. This enables the operator to draw a rough estimate of the contour and let the program refine the shape to a more accurate location on the endocardial boundary. Successive frames in the sequence are then tracked automatically using the contour found at the previous time frame to match the current one. Tracking operates at a speed of around 10 frames per s and endocardial tracking time for a 3 beat sequence was typically around 3 min (512 Mb-RAM 1.2-GHz Pentium-II Windows-based system). The endocardium is tracked over the entire sequence. Only end-systolic and end-diastolic frames are used to compute EF (biplane Simpson's method, average of the 3 beats), whereas the whole time-series dataset is used for automatic assessment of RWM. For this purpose, the amount
of radial shortening is calculated for every contour point in each segment, and normalized by its end-diastolic distance to a LV floating centroid [14]. Therefore, radial shortening is a dimensionless parameter that represents the mean shortening for each of the 16 segments. Time to peak radial shortening was expressed normalized to the cardiac cycle. Two blinded observers manually segmented visually identified end-systolic and end-diastolic frames for study I-B. For study II, reference endocardial position was traced visually frame-by-frame for the whole sequence, before automatic tracking. Subjective reading of RWM of stress-echo digital images was performed by two level-III readers [15] using a 4level scoring scale. Discrepancies were solved by consensus. 1.5. Statistical analysis Continuous variables are expressed as mean ± standard deviation. Agreement analysis between techniques (averaged values of 2 observers for study I-A) and within and between
Fig. 2. Validation analysis of the tracking system against MR (study I). Panels A–C: Linear regression plots between values provided by MR (horizontal axis) and tracking of echocardiographic sequences (vertical axis) for measuring EDV (A), ESV (B) and EF (C); the thick continuous line represents the linear fitting to the data, whereas the dotted thin line represents the identity line. Panels D–F: Bland–Altman plots for the agreement between techniques for these same 3 parameters.
J. Bermejo et al. / International Journal of Cardiology 124 (2008) 47–56
51
Table 2 Comparative analysis of the accuracy tracking system and manual tracing of the endocardial boundary using MR as reference for patients in group I Manual tracing
End-diastolic volume End-systolic volume Ejection fraction
Absolute error
Relative error (%)
32 ± 34 ml 17 ± 28 ml 0 ± 9%
23 ± 22 22 ± 31 2 ± 21
Tracking system R
Ric
0.86 0.91 0.80
0.75 0.85 0.80
Absolute error
Relative error (%)
31 ± 30 ml 19 ± 26 ml − 2 ± 9%
21 ± 21 27 ± 33 − 5 ± 17
R
Ric
0.89 0.92 0.83
0.78 0.86 0.82
R: Pearson's correlation coefficient. Ric: intraclass correlation coefficient.
operators (study I-B) was assessed by Bland–Altman analysis, as well as Pearson (R) and intraclass (Ric) correlation coefficients. For study II, the agreement between endocardial areas among all frames in the sequence was compared with manually traced values. Also, average and Hausdorff distances between tracked contour and manually segmented contour points were calculated for each frame. The Hausdorff distance measures the maximal minimum distance between manually and automatically tracked contours. Repeated-measures analysis of variance followed by Tukey contrasts were used to assess differences in quantitative parameters during the stress-echo exams. Logistic regression and receiver–operatorcharacteristic (ROC) curve analysis were used to assess the diagnostic accuracy of the system to account for RWM abnormalities. A p value b.05 was considered significant.
20%, 27 ± 33%, and − 4 ± 18%, respectively. Corresponding Ric values were 0.80, 0.87, and 0.82. Agreement between the echo tracking system and MR was better for the group I-B than for the group I-A both for measuring ESV (error = 3 ± 18% vs. 45 ± 30%, p b 0.01) and EF (error = 6 ± 16% vs. −12 ± 15%, p b 0.01). Table 2 summarizes the agreement analysis between manual tracing, and the tracking system against MR for patients in group I. As shown, the tracking system was as accurate as averaged manual segmentations for measuring ventricular volumes and EF, whereas it decreased variability (Table 3). In fact, the variability of the tracking system was slightly lower than the variability of MR.
2. Results
Analysis of a total of 3481 frames showed a regional agreement between the tracking-system and visual segmentation of a few millimeters, very close to the variability limits of visual interpretation (Table 4).
2.1. System validation against MR for measuring LV volumes and EF, and impact on variability (study I) Fig. 2 shows the agreement analysis between volumes and EF calculated by the tracked echocardiographic sequences and values obtained by MR for the total of group-I. Relative errors for the estimation of end-diastolic and end-systolic volumes (EDV and ESV, respectively), and EF were 21 ± Table 3 Analysis of the variability of the different techniques used to determine LV volumes and EF for study I-B
Intraobserver manual tracing Interobserver manual tracing Intraobserver tracking system Interobserver tracking system Intraobserver magnetic resonance Interobserver magnetic resonance
End-diastolic volume
End-systolic volume
Ejection fraction
Variability Ric (%)
Variability Ric (%)
Variability Ric (%)
2±6
0.98 0 ± 14
0.99 4 ± 18
0.96
6±9
0.95 1 ± 10
0.99 4 ± 17
0.94
1±4
0.99 1 ± 7
1.00 1 ± 5
0.99
0 ± 10
0.99 2 ± 9
0.99 1 ± 8
0.98
0±6
0.99 2 ± 5
1.00 2 ± 5
0.99
8±6
0.97 7 ± 11
0.99 0 ± 16
0.96
Abbreviations as in Table 2.
2.2. Accuracy of the regional identification of the endocardial boundary for the full cardiac cycle (study II)
2.3. Indices of regional wall-motion in patients undergoing stress echocardiography (study III) Images of all patients undergoing the stress-echo (study III) were suitable for analysis, except for two of them (96%). For the remaining 50 patients, apical views were found appropriate for analysis during the three phases, whereas short-axis views were found impossible to track in 11 patients (all phases). A total of 2502 myocardial segments were analyzed, of which 1954 were judged to be normokinetic (78%), 234 hypokinetic (9%), 293 (12%) akinetic, and 21 dyskinetic (1%). The kappa value for the agreement between
Table 4 Results of the regional wall motion agreement analysis (study II)
Manual to manual Tracker to manual Overall Short-axis 4 Chamber 2 Chamber Tracker to tracker
Frames (n)
Average distance (mm)
Hausdorff distance (mm)
1060
1.35 ± 0.68
4.13 ± 2.22
3481 540 1572 1369 1060
2.17 ± 1.09 1.87 ± 0.99 2.26 ± 1.30 2.20 ± 0.80 1.26 ± 0.59
6.63 ± 3.18 5.25 ± 2.43 6.75 ± 3.62 7.04 ± 2.73 4.82 ± 2.46
52
J. Bermejo et al. / International Journal of Cardiology 124 (2008) 47–56
experts readers for the interpretation of RWM analysis was 0.61. The tracking system accurately discriminated segments showing an abnormal wall motion of patients undergoing a stress-echo examination (Fig. 3). Compared to normal segments, myocardial segments judged as showing abnormal RWM evidenced lower values of radial shortening and reached their peak shortening later during the cardiac cycle. Mean values (95% CIs) of radial shortening were 0.27 (0.25–0.30), 0.33 (0.30–0.36), and 0.18 (0.15–0.21) for basal, mid, and apical normokinetic segments, respectively.
Thus, significant heterogeneity was observed among the 16 myocardial segments (p b .0001), due to a lower fractional radial shortening in the apical segments. The area under the ROC-curve for fractional radial shortening to predict a RWM abnormality was 0.84 (Fig. 3), and the odds-ratio for presenting an RWM abnormality was 0.31 (95% confidence interval: 0.28 to 0.35) per 10% increase in fractional radial shortening (n = 2502 segments; p b .0001). When the model was adjusted to allow for different thresholds in different segments, the area under the ROC-curve increased to 0.87. In normal subjects, dobutamine infusion increased radial
Fig. 3. Quantitative analysis of regional wall motion during stress-echo examinations (study III). Panels A and B: Values of radial shortening (A) and time to peak radial shortening (B) according to the subjective reading of regional wall motion in the 2502 myocardial segments analyzed. Panels C and D: Values of radial shortening (C) and time-to-peak radial shortening (D) according to the dobutamine dose being infused in the 1954 normokinetic segments. Norm: Normal; Hypo: hypoquinetic; A-Dys: pooled akinetic and dyskinetic segments. Base: Baseline; Low: low-dose dobutamine infusion; Peak: peak-dose dobutamine infusion. ⁎: Significantly different from normokinetic or baseline; †: significantly different from low-dose. Boxplots represent the 50th (white lines), 25th, and 75th percentile values, as well as the limits of the distribution (whiskers). The shaded zones account the 95% confidence interval for the median. Panel E: Receiver operator-characteristic (ROC) curve analysis for semi-automatically predicting a RWM abnormality, based on different cut-off values of regional radial shortening.
J. Bermejo et al. / International Journal of Cardiology 124 (2008) 47–56
53
shortening and shortening velocity only at low dobutamine doses (Fig. 3). Examples of the applicability of the system to quantify three different tress-echo responses is shown in Fig. 4. 3. Discussion 3.1. Semiautomatic analysis of echocardiographic sequences
Fig. 4. Application of the tracking system to the quantitative analysis of stress-echocardiograms. Three different response patterns are illustrated. Fractional radial shortening is shown for the six segments examined in the apical 2 chamber views from 3 different patients. Panel A: Normal response, negative result. Panel B: Abnormal ischemic response in the apical territory. Notice the absence of increase in radial shortening at the apical segments during low dobutamine dose infusion and further decrease at peak dose. This patient had an angiographically proven 95% stenosis of the mid left anterior descending coronary artery. Panel C: Abnormal biphasic response in a patient with ischemic dilated cardiomyopathy. There is a remarkable increase of radial shortening in the apical segments during low-dose dobutamine infusion followed by a subsequent decrease at peak dobutamine dose. This patient had 3-vessel coronary artery disease, his baseline ejection fraction was 28% and increased to 38% after coronary revascularization. Inf: Inferior segments; Ant: anterior segments.
Since the introduction of cross-sectional two-dimensional echocardiography, a number of attempts have been made to develop a method for quantifying LV systolic function by semi-automatically identifying the endocardial boundary [16,17]. Acoustic quantification (AQ, Philips Medical Systems) generated most expectations, because it was the first tool to become available on the ultrasound scanners [18]. Due to its ability to work throughout the full cardiac cycle, this system provided measurements of both global and regional systolic and diastolic ventricular function [19]. Recently, this system has been adapted to make it suitable for working on “white-blood” contrast images [5,6,20]. However, the widespread application of acoustic quantification has been limited because of the need of a high degree of userdependent image-parameter optimization. Technically, acoustic quantification algorithms are based on a simple threshold-based segmentation of cross-sectional ultrasound radiofrequency data, without incorporating any spatial or temporal knowledge of ventricular shape or motion [18]. However, during the last decade, a number of computer medical image processing techniques have allowed significant improvements in this field. Also, the introduction of digital storage has provided the possibility of computer post-processing echocardiographic sequences without any loss in image quality. Consequently, well established complex pre-segmentation and filtering techniques, as well as contour-based (“snake”) algorithms of spline fitting can now be incorporated to the analysis of echocardiographic images. These new techniques allow automatic systems to incorporate geometrical and temporal constraints, resulting in a more accurate following of the endocardial border. A number of authors have demonstrated the value of contrast agents to improve the identification of the bloodendocardial boundary, both at baseline and during stress echocardiography [2]. Furthermore, LV volumes and EF are more accurately estimated by echocardiography when contrast agents are used to improve image quality, both in patients with severe ventricular remodeling [21] and in consecutive patients referred for a standard echocardiogram [22]. On this basis, we hypothesized that contrast-enhanced echocardiographic sequences currently provide the highest quality ultrasound images for post-processing. Although the type of image processing algorithms used in our study have been recently applied to contrast sequences [7], to our knowledge this is the first study to clinically assess the applicability and accuracy of a contrast-based system suitable
54
J. Bermejo et al. / International Journal of Cardiology 124 (2008) 47–56
for the combined analysis of global and regional LV systolic function. 3.2. Quantification of LV volumes and EF For the purpose of measuring LV volumes and EF, the results of our study demonstrate agreement limits against MR similar to those observed by subjective manual segmentation. It is remarkable that the limits of agreement found for the tracking system (Fig. 2-F) are virtually identical to values reported in a recent multicentric clinical trial designed to assess the accuracy of manually traced contrast echocardiography against MR [23]. Automatic systems are not intended to increase the accuracy of contrast-based methods. Instead, these systems are designed to reduce observer variability, providing more reproducible results among observers and institutions and in less time. Accurate and reproducible values of EF have become crucial to take a number of therapeutic decisions in patients with impaired LV systolic function, such as indicating vasodilator therapy [24], or a cardioverter–defibrillator [25]. A semi-automatic tracking system could be particularly useful when these decisions are being considered. Additionally, a reduction in technical variability is particularly appreciated for the purpose of reducing the sample size of LV remodeling clinical trials [25]. The results of our study suggests that the objective of increasing reliability above manual segmentation techniques [23] is achieved by the system. In agreement with previous manual and semi-automatic studies using contrast ventricular opacification, the findings of our study show that LV volumes are smaller than those measured by MR [7,22,23]. Because the under-estimation of LV volumes obtained using the tracking system was very similar to the degree observed using manual segmentation, we believe this finding is most probably due to problems related to image acquisition such as apical foreshortening, contrast-induced attenuation or apical destruction of microbubbles. Additionally, inaccuracy of the Simpson's method to account for true volumes in the presence of markedly abnormal ventricular shapes may be responsible for this bias. However interestingly, semi-automatic tracking systems can also be adapted to work on ultrasound volumetric datasets to obtain a full 3D + T tracking of the LV [26]. 3.3. Quantification of the extent and temporal sequence of myocardial contraction The issue of subjectivity is recognized as the major limitation of stress echocardiography [4]. In consequence, extensive training is required for interpretation of stress echocardiograms. A previous study has demonstrated that quantitative measurements of RWM may increase the accuracy of the technique, when image quality is suitable [19]. Before testing the applicability of the tracking system, we performed a comprehensive analysis of the regional accuracy
of the system, which demonstrated good concordance with manual segmentations (study II). Furthermore, study III showed the applicability of tracking systems for the purpose of stress echo. A simply derived parameter of RWM such as regional radial fractional shortening quantitatively identified the abnormalities found by expert readers. Also, the system allowed to characterize the effects of dobutamine on RWM. Interestingly, in patients showing normal subjective regional wall motion, no further increase in regional shortening was observed at dobutamine doses higher than 10 μg/kg/min, a finding in accordance with MR studies [27]. Also, our observation of significant heterogeneity among segments in the extent of endocardial excursion has been demonstrated using this technique [27,28]. This finding supports using different cut-off values to identify RWM abnormalities. Using this approach, we obtained excellent discrimination accuracy for this purpose (area under the ROC curve = 0.87). Tracking systems provide quantitative measurements not only of the extent, but also of the timing of myocardial excursion. Noticeably, increased sensitivity of stress examinations to detect myocardial ischemia has been demonstrated when temporal information is incorporated to the diagnostic algorithms [29,30]. Further large-scale clinical research should provide the physiological values of RWM, and eventually assess the final impact of semi-automatic techniques to increase the efficacy of stress-echo examinations. A quantitative analysis of RWM is potentially useful for other applications as well. Recently, the potential of characterizing the temporal course of endocardial wall motion by frequency-domain analysis of radionuclide images has shown to be useful for assessing the extent of LV dyssynchrony [31]. Because RWM is most frequently assessed using echocardiography in the clinical setting, a similar type of analysis suitable for echocardiographic examinations should be desirable. 3.4. Study limitations The agreement observed between MR and the tracking system was different between groups I-A and I-B. We believe that a number of issues can account for this finding, probably related to patient selection, the type of scanner image modalities or the administration method for the contrast agent. The confidence intervals of validation results are usually wider in multicentric than single-center reliability studies, and very close to our results using the tracking system [23]. Furthermore, agreement against MR was almost identical for the tracking system and for averaged manual tracings of an experienced observed; the system added the advantage of reducing variability. To test the system in stressecho, we used conventional RWM visual scoring as reference. Although an independent reference technique of RWM could be theoretically more robust, we believe that using visual assessment as reference is the only applicable tool for processing identical sequences as processed using the tracker system. Also, a recent study has demonstrated that
J. Bermejo et al. / International Journal of Cardiology 124 (2008) 47–56
subjective RWM reading of contrast-echo sequences is, at least, as reliable as any alternative imaging technique [32]. We implemented a floating centroid to measure radial shortening as the quantitative parameter of RWM. Further investigation should clarify whether more complex parameters such as centerline methods may hypothetically increase the sensitivity of the system [33]. Future clinical trials using an alternative imaging method as reference should clarify the final clinical value of the system to detect ischemia and/or viability in a per patient analysis. After ending patient enrolment, sulphur hexafluoride has recently been recommended not to be used in patients immediately following an acute coronary syndrome, due to safety issues. 4. Conclusions The present study demonstrates that a robust quantification of global and regional systolic ventricular function can be obtained by digital processing contrast echocardiographic sequences. Values of LV volumes, ejection fraction, and regional endocardial shortening obtained using these tools adequately correlate with currently available reference methods such as magnetic resonance. Furthermore, semiautomatic systems running on contrast-echo sequences are able to reduce the variability of echocardiography for obtaining quantitative measurements of LV volume. Readily applicable to baseline and stress-echo studies, the instantaneous blood-endocardial boundary detected by the system closely matches the visually identified border. Quantitative values of regional radial shortening provided by the system accurately match subjectively determined RWM. Thus, endocardial tracking techniques increase the reliability of echocardiography when used to assess global and regional LV systolic function. References [1] Schiller NB, Shah P, Crawford M, et al. Recommendations for quantitation of the left ventricle by two-dimensional echocardiography. American Society of Echocardiography Committee on Standards, Subcommittee on Quantitation of Two-Dimensional Echocardiograms. J Am Soc Echocardiogr 1989;2:358–67. [2] Porter TR, Xie F, Kricsfeld A, et al. Improved endocardial border resolution during dobutamine stress echocardiography with intravenous sonicated dextrose albumin. J Am Coll Cardiol 1994;23: 1440–3. [3] Crouse LJ, Cheirif J, Hanly DE, et al. Opacification and border delineation improvement in patients with suboptimal endocardial border definition in routine echocardiography: results of the Phase III Albunex Multicenter Trial. J Am Coll Cardiol 1993;22:1494–500. [4] Hoffmann R, Lethen H, Marwick T, et al. Analysis of interinstitutional observer agreement in interpretation of dobutamine stress echocardiograms. J Am Coll Cardiol 1996;27:330–6. [5] Spencer KT, Bednarz J, Mor-Avi V, et al. Automated endocardial border detection and evaluation of left ventricular function from contrast-enhanced images using modified acoustic quantification. J Am Soc Echocardiogr 2002;15:777–81. [6] Mor-Avi V, Bednarz J, Weinert L, et al. Power Doppler imaging as a basis for automated endocardial border detection during left ventricular contrast enhancement. Echocardiography 2000;17:529–37.
55
[7] Morales MA, Positano V, Lombardi M, et al. Semiautomatic detection of left ventricular contours in contrast-enhanced echocardiographic images: comparison with magnetic resonance imaging. J Am Soc Echocardiogr 2004;17:876–82. [8] Pellikka PA, Roger VL, Oh JK, et al. Stress echocardiography. Part II. Dobutamine stress echocardiography: techniques, implementation, clinical applications, and correlations. Mayo Clin Proc 1995;70:16–27. [9] Nurbekov Y, Bermejo J, Mulet-Parada M, et al. Endocardial tracking from two-dimensional integrated-backscatter echocardiographic sequences. Initial validation against manual segmentation and cineMRI. Circulation 2001;104(Suppl II):653 [Abstract]. [10] Kelion AD, Clarke NRA, Mulet M, et al. A novel automated endocardial tracker: comparison with acoustic quantification. Eur Heart J 2001;22:572 [(Abstract Suppl), Abstract]. [11] Mulet-Parada M, Noble JA. 2D + T acoustic boundary detection in echocardiography. Med Image Anal 2000;4:21–30. [12] Cootes TF, Taylor CJ, Cooper DH, et al. Active shape models — their training and application. CVGIP Image Underst 1994;61:38–59. [13] Mulet-Parada, M. Intensity independent feature extraction and tracking in echocardiographic sequences. PhD Thesis. University of Oxford. Available at: http://www.robots.ox.ac.uk/~mvl/htdocs/publications/ publications/miguel_thesis00.html, July 2005. [14] Wiske PS, Pearlman JD, Hogan RD, et al. Echocardiographic definition of the left ventricular centroid. II. Determination of the optimal centroid during systole in normal and infarcted hearts. J Am Coll Cardiol 1990;16:993–9. [15] Recommendations for continuous quality improvement in echocardiography. American Society of Echocardiography. J Am Soc Echocardiogr 1995;8:S1–S28. [16] Buda AJ, Delp EJ, Meyer CR, et al. Automatic computer processing of digital 2-dimensional echocardiograms. Am J Cardiol 1983;52: 384–389. [17] Zwehl W, Levy R, Garcia E, et al. Validation of a computerized edge detection algorithm for quantitative two-dimensional echocardiography. Circulation 1983;68:1127–35. [18] Perez JE, Waggoner AD, Barzilai B, et al. On-line assessment of ventricular function by automatic boundary detection and ultrasonic backscatter imaging. J Am Coll Cardiol 1992;19:313–20. [19] Koch R, Lang RM, Garcia MJ, et al. Objective evaluation of regional left ventricular wall motion during dobutamine stress echocardiographic studies using segmental analysis of color kinesis images. J Am Coll Cardiol 1999;34:409–19. [20] Caiani EG, Lang RM, DeCara J, et al. Objective assessment of left ventricular wall motion from contrast-enhanced power modulation images. J Am Soc Echocardiogr 2002;15:118–28. [21] Thomson HL, Basmadjian AJ, Rainbird AJ, et al. Contrast echocardiography improves the accuracy and reproducibility of left ventricular remodeling measurements: a prospective, randomly assigned, blinded study. J Am Coll Cardiol 2001;38:867–75. [22] Malm S, Frigstad S, Sagberg E, et al. Accurate and reproducible measurement of left ventricular volume and ejection fraction by contrast echocardiography: a comparison with magnetic resonance imaging. J Am Coll Cardiol 2004;44:1030–5. [23] Hoffmann R, von Bardeleben S, ten Cate F, et al. Assessment of systolic left ventricular function: a multi-centre comparison of cineventriculography, cardiac magnetic resonance imaging, unenhanced and contrast-enhanced echocardiography. Eur Heart J 2005;26:607–16. [24] The SOLVD Investigators. Effect of enalapril on survival in patients with reduced left ventricular ejection fractions and congestive heart failure. N Engl J Med 1991;325:293–302. [25] Bellenger NG, Davies LC, Francis JM, et al. Reduction in sample size for studies of remodeling in heart failure by the use of cardiovascular magnetic resonance. J Cardiovasc Magn Reson 2000;2:271–8. [26] Bermejo J, Lang RM, Odreman R, et al. Four-dimensional surfacetracking of contrast-enhanced transthoracic biplane echocardiograms: a new tool for quantitative assessment of LV function. J Am Coll Cardiol 2003;41:437A [Abstract].
56
J. Bermejo et al. / International Journal of Cardiology 124 (2008) 47–56
[27] Scott CH, Sutton MS, Gusani N, et al. Effect of dobutamine on regional left ventricular function measured by tagged magnetic resonance imaging in normal subjects. Am J Cardiol 1999;83:412–7. [28] van Rugge FP, Holman ER, van der Wall EE, et al. Quantitation of global and regional left ventricular function by cine magnetic resonance imaging during dobutamine stress in normal human subjects. Eur Heart J 1993;14:456–63. [29] Mor-Avi V, Collins KA, Korcarz CE, et al. Detection of regional temporal abnormalities in left ventricular function during acute myocardial ischemia. Am J Physiol Heart Circ Physiol 2001;280:H1770–81. [30] Voigt JU, Nixdorff U, Bogdan R, et al. Comparison of deformation imaging and velocity imaging for detecting regional inducible ischaemia during dobutamine stress echocardiography. Eur Heart J 2004;25:1517–25.
[31] Fauchier L, Marie O, Casset-Senon D, et al. Interventricular and intraventricular dyssynchrony in idiopathic dilated cardiomyopathy: a prognostic study with Fourier phase analysis of radionuclide angioscintigraphy. J Am Coll Cardiol 2002;40:2022–3030. [32] Hoffmann R, von Bardeleben S, Kasprzak JD, et al. Analysis of regional left ventricular function by cineventriculography, cardiac magnetic resonance imaging, and unenhanced and contrast-enhanced echocardiography: a multicenter comparison of methods. J Am Coll Cardiol 2006;47:121–8. [33] Nidorf SM, Wieyman AE. Left ventricle II: quantification of segmental dysfunction. In: Weyman AE, editor. Pinciples and practice of echocardiodiography. Second ed. Pennsylvania: Lea & Febiger; 1994. p. 625–56.
