Segmental Analysis of Color Kinesis Images
New Method for Quantification of the Magnitude and Timing of Endocardial Motion During Left Ventricular Systole and Diastole
Background We describe a method for objective assessment of left ventricular (LV) endocardial wall motion based on Color Kinesis, a new echocardiographic technique that color-encodes pixel transitions between blood and myocardial tissue.
Methods and Results We developed a software that analyzes Color Kinesis images and provides quantitative indices of magnitude and timing of regional endocardial motion. Images obtained in 12 normal subjects were used to evaluate the variability in each index. Esmolol, dobutamine, and atropine were used to track variations in LV function in 14 subjects. Objective evaluation of wall motion was tested in 20 patients undergoing dobutamine stress testing. Regional fractional area change, displacement, and radial shortening were displayed as histograms and time curves. Global function was assessed by calculating magnitude and timing of peak ejection or filling rates and mean time of ejection or filling. Patterns of endocardial motion were consistent between normal subjects. Fractional area change and peak ejection rate decreased with esmolol and increased with dobutamine. Time to peak ejection and mean time of contraction were prolonged with esmolol and shortened with dobutamine. Using atropine, we proved that our findings with dobutamine were not secondary to its chronotropic effects. Dobutamine induced regional wall motion abnormalities in 10 patients in 38 segments diagnosed conventionally. Segmental analysis detected abnormalities in 36 of these 38 segments and in an additional 5 of 322 segments.
Conclusions Analysis of Color Kinesis images allows fast, objective, and automated evaluation of regional wall motion sensitively enough to evaluate clinical dobutamine stress data. This method has significant potential in the diagnosis of myocardial ischemia.
Two-dimensional echocardiography is currently the most widely used noninvasive imaging modality for the evaluation of regional left ventricular (LV) wall motion. However, conventional assessment of wall motion, based on visual interpretation of endocardial excursion and myocardial thickening, is subjective and experience-dependent. Previous attempts to quantify LV wall motion have been based primarily on manual, time-consuming, frame-by-frame tracing of the endocardial border.
To facilitate a more objective evaluation of LV endocardial motion, Color Kinesis, a new technique based on acoustic quantification,1 2 3 4 5 has recently been developed and incorporated into a commercial ultrasound system (Hewlett-Packard). This technique compares backscatter values between successive acoustic frames and generates color overlays in which different colors are used to encode pixel transitions between blood and myocardial tissue. These overlays are superimposed on two-dimensional images and updated in real time on a frame-by-frame basis by addition of one color at a time. Thus, end-systolic or end-diastolic overlays provide in a single frame an integrated display of the timing and magnitude of endocardial motion of the most recent heartbeat (Fig 1⇓). With appropriate methods of analysis, Color Kinesis may provide objective evaluation of endocardial motion on a global as well as regional basis.
Accordingly, the aims of this study were (1) to develop a method for automated quantitative analysis geared toward providing easily interpretable indices of both magnitude and timing of LV endocardial motion during ejection and filling, (2) to test the feasibility of this approach by inducing predictable alterations in the magnitude and timing of endocardial motion with pharmacological interventions, and (3) to determine the feasibility of using this methodology for objective diagnosis of regional wall motion abnormalities in patients undergoing dobutamine stress testing.
Study Population and Design
In protocol 1, Color Kinesis images were initially obtained in 12 normal subjects (3 women, 9 men; mean age, 34±5 years) selected on the basis of having high-quality echocardiographic images. These images were used to test the feasibility of a newly developed software for segmental analysis of Color Kinesis images and evaluate the intersegment and intersubject variability of each measured parameter. Subsequently, two separate protocols were carried out to study the ability of our analysis to reflect variations in LV systolic and diastolic function induced by different pharmacological agents in normal subjects.
In protocol 2, end-systolic and end-diastolic Color Kinesis images were obtained in 7 normal subjects (5 men, 2 women; age, 33±3 years) under control conditions and during intravenous infusions of the cardioselective β-blocker esmolol and β-agonist dobutamine. Esmolol was administered as a 500-μg/kg bolus followed by continuous infusion of 50, 100, 150, and 200 μg·kg−1·min−1. Before each change in infusion dose, the subject received a repeated bolus of 500 μg/kg. Images were acquired after 15 minutes of infusion of the highest dose of esmolol. Thirty minutes after termination of esmolol infusion, dobutamine was infused (15 μg·kg−1·min−1), and an additional set of images was acquired.
Protocol 3 was designed to determine the relative contributions of augmented contractility versus increased heart rate toward the variations in Color Kinesis data observed with dobutamine. This protocol was conducted in 7 normal subjects (4 men, 3 women; age, 32±6 years). After data acquisition under control conditions, end-systolic Color Kinesis images were obtained with dobutamine at 5, 10, and 15 μg·kg−1·min−1. To match the heart rate noted with the highest dose of dobutamine, atropine was administered intravenously in increments of 0.2 mg (up to 1.2 mg) 30 minutes after termination of dobutamine infusion, and data acquisition was repeated.
Protocol 4 was designed to test the feasibility of segmental analysis of Color Kinesis images for objective diagnosis of regional wall motion abnormalities induced during dobutamine stress testing. Two-dimensional images were obtained in 20 randomly selected patients (14 women, 6 men; age, 62±2 years) undergoing a standard protocol for dobutamine stress echocardiography.6 Images were acquired under baseline conditions, low and peak doses of dobutamine, and during recovery. Simultaneously, Color Kinesis was activated to color encode systolic endocardial motion. Image sequences with Color Kinesis overlays were saved in digital format on optical disks for subsequent off-line analysis.
Ultrasound imaging was performed with a 2.5- or 3.5-MHz transducer (SONOS 2500, Hewlett-Packard) with the subject in the left lateral decubitus position. In protocols 1, 2, and 3, parasternal short-axis views at the midpapillary level and apical four-chamber views were obtained during end expiration. In patients undergoing dobutamine stress testing, images were also obtained in the parasternal long-axis and apical two-chamber views. In this protocol, images were acquired with a commercial software package for stress echocardiography in a quad-screen format.
After image quality was optimized, the acoustic quantification system was activated. Gain controls, including total and lateral gain and time gain compensation, were adjusted to optimize tracking of the endocardial boundary,7 and a region of interest surrounding the LV cavity was defined. Color Kinesis was then activated to color-encode endocardial motion within this region of interest. Color encoding of endocardial motion during systole was triggered by the R wave of the ECG. Duration of systole, T, was automatically calculated by the imaging system from the following empirical equation: where T is expressed in ms and HR is heart rate in beats per second.8 For each imaging plane, two systolic sequences of Color Kinesis images were obtained. Subsequently, Color Kinesis was switched to color-encode endocardial motion during diastole. The timing of color encoding was set to start at end systole and terminate 19 frames (19×33=627 ms) later or at the ensuing R wave, whichever occurred first. Since the empirical formula in Equation 1 may become inaccurate, in particular under β-adrenergic stimulation, manual adjustments in the time settings of color encoding were performed when necessary. This was done so as to ensure that the initiation of color encoding coincided with the first video frame in which outward endocardial motion was noted. For each view, two diastolic image sequences were acquired. All images were stored in a digital format on optical disk for off-line analysis.
Digital images were reviewed off-line with the continuous-loop review mode of the echocardiographic system to ensure accurate tracking of the endocardial boundary. The end-systolic or end-diastolic frames were then analyzed with custom-designed image processing software written for a personal computer. Images in each view were partitioned on the basis of standard segmentation models corresponding to the coronary perfusion territories9 (Fig 2⇓).
In the short-axis view, the segmentation originated from the LV end-systolic cavity centroid. The zero-degree line was defined by the centroid and a manually determined anatomic landmark represented by the junction between the right ventricular posterior wall endocardium and the interventricular septum10 11 12 13 14 (Fig 2A⇑). The left ventricle was then divided into six 60° wedge-shaped sectors. In the apical two- and four-chamber views, each image was initially divided into two sections separated by a line (defined as the long axis) connecting the distal apical endocardium to the end-systolic cavity centroid. In each section, a wedge-shaped sector was defined between the long axis and a line connecting the centroid with the mitral valve leaflet. This scheme excluded mitral valve motion from the analysis.15 Each sector was then further divided into three equiangled sectors, resulting in a total of six sectors originating from the cavity centroid (Fig 2B⇑ and 2C⇑).
In each segment, pixels of each color and pixels marked as blood were counted, and pixel counts were averaged for each pair of end-systolic and end-diastolic images obtained in each view. Pixel counts were subsequently used for the evaluation of regional endocardial motion. The magnitude of endocardial motion was expressed in terms of incremental fractional area change (in percent of segmental and global end-diastolic area), endocardial wall displacement (in millimeters), and fractional radial shortening (in percent of end-diastolic radial dimension). The temporal characteristics of regional endocardial motion were assessed by use of ejection and filling rates as a function of time, time curves reflecting integrated fractional area change, and integrated fractional radial shortening, as well as mean time of ejection and filling. Analysis techniques used to obtain these parameters are described in detail in the “Appendix.”
Automated Detection of Regional Wall Motion Abnormalities
In protocol 4, dobutamine-induced regional wall motion abnormalities were identified in segments that displayed reduced magnitude of endocardial motion relative to baseline conditions. Abnormality was defined as reduction in regional fractional area change of >40% of the baseline value. This threshold for automated detection was based on the results of our previous study,16 in which we observed normal variability of up to 20% in data obtained at rest.
Visual Interpretation of Regional Wall Motion
Conventional visual interpretation of two-dimensional images was performed by an experienced reader of dobutamine stress echocardiograms who was unaware of the results of Color Kinesis analysis. The interpretation was based in each view on segmentation schemes recommended by the American Society of Echocardiography.9 For each patient, endocardial motion in each segment was examined visually and judged as either normally responding to dobutamine or hypokinetic as a result of the dobutamine infusion. This classification of segments as normal or abnormal was compared with that based on the automated detection of dobutamine-induced regional wall motion abnormalities on a segment-by-segment basis.
Data obtained in protocols 1 through 3 were averaged and composite displays of all measured parameters generated. To assess the intersubject variability for each measured parameter in normal subjects, the SD divided by the mean (averaged for all segments in each imaging plane) was computed. In protocol 2, the differences between each drug and control conditions were tested by Student’s t test. In protocol 3, ANOVA was used to test the differences between the different phases and control conditions. Differences were considered significant for P<.05. In protocol 4, statistics of agreement and disagreement between the visual interpretation of two-dimensional images and the automated detection of regional wall motion abnormalities were obtained in terms of underscoring and overscoring by the automated detection compared with the conventional methodology.
Protocol 1: Normal Subjects Under Control Conditions
Magnitude of Endocardial Motion
Figs 3 through 5⇓⇓⇓ show the results of our measurements of magnitude of endocardial motion during LV systole and diastole. Fig 3⇓ shows stacked color-encoded histograms reflecting the incremental area change in each segment in percent of segmental end-diastolic area during systole (left) and diastole (right) in the short-axis (top) and apical four-chamber (bottom) views. Color hues correspond to those used in Color Kinesis images. In each view, the patterns of endocardial motion during systole and diastole were similar. These patterns were more uniform between segments in the short-axis view compared with the apical four-chamber view, in which reduced endocardial motion was noted in the apical segments compared with other segments.
Incremental area change in percent of global (rather than segmental) LV end-diastolic area also showed similar patterns of motion during contraction (Fig 4⇑, left) and relaxation (right) in each view but with more intersegmental variability. In the short-axis view, a peak was noted in the posterior segment (Fig 4⇑, top), and in the apical four-chamber view, basal and apical segments showed augmented motion compared with midlateral and midseptal segments (bottom).
Endocardial wall displacement showed patterns of contraction and relaxation similar to those in Fig 4⇑ but with less intersegmental variability. Another difference between these data sets was the increased contribution of late colors in systole and early colors in diastole to overall wall displacement. The patterns of fractional radial shortening were almost identical to those of segmental fractional area change (Fig 3⇑). Fractional radial shortening had more uniform patterns of contraction and relaxation compared with wall displacement.
The intersubject variability for each measured parameter indicated that the short-axis view provided more consistent measurements during both contraction and relaxation than the apical four-chamber view (Table 1⇓, columns 1 and 2 versus columns 3 and 4). In both views, the lowest variability was obtained when endocardial motion was measured with regional fractional area change.
Timing of Endocardial Motion
Figs 5 through 8⇑⇓⇓⇓ present the results of the quantification of the temporal aspects of LV systole and diastole. Time histograms reflecting the fractional area change versus time showed marked differences between contraction and relaxation while remaining similar between views in both systole and diastole (Fig 5⇑). The mean time of contraction was 135±11 ms in the short-axis and 133±9 in the apical four-chamber view. The time to peak filling rate was 143±38 and 181±27 ms from end systole in the short-axis view and the apical four-chamber view, respectively. The mean time of filling was 198±26 ms in the short-axis and 195±19 in the apical view.
Regional mean time of ejection and filling was found to be uniform in the short-axis view (Fig 6⇑, top). In contrast, more pronounced intersegmental variability was observed in the apical four-chamber view (Fig 6⇑, bottom). Integrated fractional area change time curves obtained in each subject during contraction and relaxation provided a simple display of the temporal progression of ejection and filling phases in each view (Fig 7⇑). When combined, these two curves reflected changes in LV area throughout the cardiac cycle. In the short-axis view, total systolic and diastolic fractional area changes were 66±6% and 77±7%, respectively. In contrast, in the apical four-chamber view, the corresponding values were 28±6% and 36±8%. Regional time curves reflecting integrated area change (Fig 8⇑) and radial shortening showed almost identical temporal patterns, which were uniform between segments and consistent between subjects. In most segments, ejection and filling phases (rapid filling, diastasis, and atrial filling) were clearly depicted. These regional patterns were similar to those reflecting global fractional area change (Fig 7⇑).
Fig 9⇓ shows end-systolic Color Kinesis images obtained in one subject under control conditions and during infusions of esmolol and dobutamine. In all subjects, in addition to differences in magnitude of endocardial motion, consistent differences in color distribution were observed with each pharmacological intervention. With esmolol, the relative contribution of early colors (orange and yellow) was reduced compared with control conditions, indicating decreased endocardial motion during early systole. In contrast, dobutamine resulted in increased contribution of these early colors, reflecting augmented endocardial motion during early systole. These differences were objectively demonstrated by quantitative segmental analysis. Fig 9⇓ (bottom) presents the summary of results obtained in protocol 2 as stacked color histograms reflecting regional incremental fractional area change throughout systole in the short-axis view. Analysis of end-diastolic images showed similar variations.
Time histograms reflecting ejection and filling rates versus time demonstrated the drug-induced differences in overall systolic and diastolic function (Fig 10⇓). These differences are summarized in Table 2⇓. Compared with control conditions, global LV fractional area change and filling fraction (total area under the time histogram) and peak ejection and filling rates decreased with esmolol and increased with dobutamine. Time to peak ejection and filling rates and mean time of contraction and filling were prolonged with esmolol and shortened with dobutamine compared with baseline. The regional mean time of contraction and filling also reflected the expected pharmacological effects but on a regional basis.
Regional time curves reflected the drug-induced variations in the temporal progression of LV contraction and relaxation (Fig 11⇓). The slope of the initial portion of the curves corresponding to early systole and diastole decreased with esmolol and increased with dobutamine compared with baseline. For example, in systole, at 50% contraction time, 61±2% contraction was completed under control conditions, 54±4% with esmolol (P<.001), and 78±5% with dobutamine (P<.001).
Fig 12⇓ compares the effects of dobutamine on the patterns of systolic endocardial motion with those of atropine at matched heart rates. Compared with control conditions (top left), the relative contribution of early colors in the histograms consistently grew with dobutamine (top center), reflecting a progressive increase in endocardial motion during early systole. With atropine (top right), the color distribution was similar to that observed at baseline despite the differences in heart rate (Table 3⇓). Regional time curves (bottom) reflected the dobutamine-induced variations in the temporal progression of regional LV contraction. Compared with control conditions (bottom left), the slope of the initial portion of the curves consistently increased with dobutamine (bottom center), reflecting again the augmented endocardial motion in early systole. The temporal progression of contraction with atropine (bottom right) was similar to that at baseline despite the significant differences in heart rate. For example, in the short-axis view, at 50% contraction time, 65±3% contraction was completed under control conditions, 88±6% with dobutamine (P<.001), and 69±4% with atropine (P=NS). Data obtained in the apical four-chamber view yielded similar results.
These effects of dobutamine were found to be dose dependent. Table 3⇑ presents a summary of results of protocol 3. Dobutamine resulted in a significant increase in fractional area change and peak ejection rate. Time to peak ejection rate and mean time of contraction shortened progressively with increasing doses of dobutamine. In contrast, with atropine, no significant differences were found in these parameters compared with baseline.
Clinical Trials: Protocol 4
Fig 13⇓ shows stacked histograms of regional fractional area change obtained in the short-axis (left), apical four-chamber (center), and two-chamber (right) views in a patient undergoing dobutamine stress under control conditions (top), low dose (middle), and peak dose (bottom). To allow objective detection of induced regional wall motion abnormalities, histograms obtained with dobutamine are superimposed on that obtained under control conditions (hatched background).
Conventional visual interpretation of the two-dimensional images obtained during dobutamine stress testing detected induced regional wall motion abnormalities in 10 of 20 patients in 38 segments (of 360 segments examined). Segmental analysis of end-systolic Color Kinesis images showed a reduced magnitude of endocardial motion compared with control conditions in 36 of these 38 segments (95%). The 2 cases of apparent failure to detect abnormalities (5%) were found to be a result of discordance between segmentations used during visual interpretation of two-dimensional images and automated analysis of end-systolic Color Kinesis images. In addition, segmental analysis of Color Kinesis showed reduced fractional area change under stress in 5 of 322 segments that were not classified as abnormal by visual interpretation (<2%). These discordantly positive detections were a result of inadequate endocardial tracking as confirmed by repeated examination of image sequences with and without Color Kinesis overlays.
Fig 14⇓ shows an example of data obtained in a patient who developed a regional wall motion abnormality during dobutamine infusion. At rest (left), regional fractional area change (top) and mean time of ejection (bottom) were normal compared with those shown in Figs 3⇑ and 6⇑. With a low dose of dobutamine (center), septal segments showed magnitude of motion slightly below the baseline level (top), and the mean time of ejection in these segments did not change, whereas in the lateral wall, the mean time was short compared with baseline (bottom). At peak dobutamine (right), the mid-septal and basal-septal segments became severely hypokinetic (top). The mean time of ejection in these segments was significantly shorter than in the other segments, which showed appropriate augmentation in response to dobutamine (bottom). Similar observations were made in 2 additional patients who developed severe hypokinesis during dobutamine infusion.
Routine echocardiographic evaluation of regional LV function is traditionally based on visual interpretation of regional myocardial thickening and endocardial motion. Quantitative analysis of LV endocardial motion has been very difficult, because it required time-consuming, off-line, frame-by-frame manual tracing of endocardial boundaries.10 11 15 17 18 19 20 The development of acoustic quantification1 2 provided automated real-time detection of the endocardial border and continuous signals reflecting global LV cross-sectional area or volume throughout the cardiac cycle,3 4 5 thus eliminating the need for the tedious manual tracing of multiple frames. However, acoustic quantification signals do not reflect regional LV function.
Color Kinesis is a new technique based on acoustic quantification, which provides automated tracking and color encoding of endocardial boundaries. Validation of Color Kinesis, as a new technique, before attempted clinical use is of primary importance. However, Color Kinesis is based on exactly the same techniques as used to differentiate myocardial tissue from blood by acoustic quantification, which has been extensively validated.3 4 5 21 22 23 24 25 26 Color Kinesis differs from acoustic quantification only in the way this information is used to display endocardial motion.
In this study, we used this modality to develop a new methodology for objective and quantitative evaluation of regional wall motion. A major advantage of this new technique is that regional wall motion can be analyzed from a single end-systolic or end-diastolic image. In this article, we described detailed methods of segmental analysis of Color Kinesis images and provided a step-by-step description of how various indices of magnitude and timing of regional endocardial motion can be obtained from these images. We evaluated the ability of this approach to assess variations in spatial and temporal patterns of endocardial motion induced by pharmacological interventions with known effects and studied the feasibility of this methodology to objectively detect regional wall motion abnormalities in patients undergoing dobutamine stress testing. A method for objective evaluation of wall motion from echocardiographic images obtained under conditions of dobutamine stress testing would be of particular clinical value, because these images are even more difficult to interpret than conventional echocardiograms.
Initially, two important questions had to be answered before individual parameters could be derived from Color Kinesis images. First, should a floating frame of reference be used to compensate for possible cardiac translation or rotation? And second, which segmentation schemes would best match endocardial motion throughout the cardiac cycle while respecting the coronary perfusion territories?
Choice of Reference Frame
Previous investigators have demonstrated the advantages of a floating over a fixed reference frame,14 17 27 whereas others did not find significant differences between the two methods.18 28 Different methods of image alignment based on floating reference have been used to correct for cardiac translation and rotation that may occur during the cardiac cycle. These methods used the centroid of either the endocardial boundary or the entire LV cavity14 15 18 or other points17 29 as stationary points for image alignment.
We chose to use a fixed reference system to analyze end-systolic and end-diastolic Color Kinesis images. This choice was made because Color Kinesis adopted a fixed reference system for tracking endocardial motion by comparing identical pixels in successive acoustic frames. In other words, a fixed frame of reference is used at the time that Color Kinesis images are created. Moreover, since each pixel in the color overlay cannot be assigned more than one value, information regarding multiple pixel transitions resulting from translation and rotation is not reflected in the end-systolic or end-diastolic color overlays. Consequently, although a floating reference frame analysis could be applied off-line to Color Kinesis images, to provide an accurate correction for translation and rotation artifacts, it would need to be performed on a frame-by-frame basis rather than by analyzing a single end-systolic or end-diastolic frame.
Schemes used to segment the left ventricle in the different views should address two different issues. First, since regional wall motion abnormalities are usually related to coronary perfusion defects, segmentation models should respect the anatomic perfusion territories. Accordingly, we followed the segmentation schemes described in the guidelines of the American Society of Echocardiography.9 Implementation of these schemes in our automated digital image analysis required detailed decisions regarding the origin of segmentation as well as the size and geometry of each segment. Segmentation of both short-axis and apical views originated from the end-systolic LV cavity area centroid, which has previously been shown to provide more reproducible data than the centroid of the LV boundary.27 In the short-axis view, it was based on a zero line connecting the centroid with the junction of the right ventricular posterior wall endocardium and the interventricular septum.10 11 12 13 14 This latter anatomic landmark was chosen because it was previously shown to be minimally affected by cardiac rotation and translation at rest.13 To minimize the variability introduced by individual differences in LV orientation in the apical views, we used two easily identifiable anatomic landmarks at the mitral annulus and one at the apical endocardium.
Another important issue related to the design of segmentation models is that the geometry of individual segments should match the regional patterns of endocardial motion and reflect as closely as possible the motion of each specific anatomic segment. To this effect, segmentation lines should ideally be parallel to regional velocity vectors in each frame. Thus, in the short-axis view, a radial segmentation scheme provided a simple solution to follow the fairly symmetrical radial motion occurring during LV contraction and relaxation. In the apical four-chamber view, endocardial motion has been described as simultaneous inward motion related to wall thickening and motion toward the apex secondary to shortening of the LV long axis,30 31 which has been described as gradually decreasing from base to apex.32 To follow this complex motion pattern, segmentation lines should be drawn almost perpendicular to the long axis with a slight inclination toward the apex, which should decrease from base to apex. This hypothetical nonconcentric segmentation model would be considerably more difficult to implement than a simple concentric segmentation similar to the one we used in the short-axis view. Therefore, we chose to use a segmentation scheme originating from a single well-defined point, the LV cavity centroid. In fact, the simple concentric and the complex nonconcentric segmentation schemes provide similar segment geometry, with the exception of the boundaries between the apical and midventricular segments.
Interpretation of Results
Normal Subjects at Rest
Given the technical ability of Color Kinesis to depict endocardial motion throughout systole or diastole by automated analysis of a single color-encoded image, we analyzed end-systolic as well as end-diastolic frames obtained in 12 normal subjects with the above-described segmentation models. The automated quantitative analyses were designed to provide different indices of magnitude and timing of endocardial motion.
Measurements of magnitude of endocardial motion. To quantify the magnitude and timing of endocardial motion, different indices, such as fractional area change in percent of global as well as regional end-diastolic area,10 11 14 15 17 18 metric displacement, and regional fractional shortening,10 14 18 19 were used. We found that each of these parameters had similar histogram morphology in systole and diastole in each view, except for the individual color distribution, which reflected temporal differences between contraction and relaxation (Figs 3 through 5⇑⇑⇑).
In both views, regional fractional area change in percent of segmental end-diastolic area (Fig 3⇑) and fractional radial shortening had more uniform patterns compared with fractional area change in percent of global end-diastolic area of the LV cavity (Fig 4⇑). This difference may be explained by the fact that normalization by a global rather than regional parameter, such as end-diastolic area of the entire LV cavity, fails to eliminate intersegmental geometrical differences. Thus, it appears that regional normalization is advantageous for detection of regional wall motion abnormalities, because it allows intersegmental comparisons regardless of geometric differences. Using regional normalization for both fractional area change and radial shortening, we confirmed the previously described reduced motion in the apical lateral segment due to the limited ability to visualize and track the apical lateral boundary.15 33 This has been attributed to lung interference, myocardial anisotropy, and/or the angle between muscle fibers and ultrasonic beam.33 34
Endocardial wall displacement that did not involve regional normalization was also found to have less intersegmental variation than the globally normalized fractional area change (Fig 4⇑). This former parameter provides an absolute measurement of regional wall motion and therefore should be less affected by variations in radial distance from the centroid during contraction or relaxation than fractional area change. However, in this study, displacement was calculated with an arc approximation for each color band, the validity of which varies from segment to segment. In particular, this geometric assumption fails in the inferior, posterior, and lateral segments in the short-axis view, because these segments contain the papillary muscles. Similarly, it does not correctly represent the geometry of the midventricular and basal segments in the apical views. It therefore seems that more specific modeling of the regional endocardial geometry would be necessary to obtain more reliable data.
Regional fractional area change in percent of segmental end-diastolic area was found not only to best reflect the uniform pattern of endocardial motion but also to provide the most consistent intersubject measurements (Table 1⇑). We believe that this variability between subjects reflects physiological differences encountered in a normal population. Thus, on the basis of the results of this study and within the limitations of the specific methods of analysis used, regional fractional area change appears to provide the most reliable information on the magnitude of regional endocardial motion.
Measurements of timing of endocardial motion. Before the development of acoustic quantification, echocardiographic assessment of the timing of endocardial motion has been extremely difficult.11 35 36 The advent of acoustic quantification allowed real-time acquisition of continuous signals reflecting global LV cross-sectional area or volume throughout the cardiac cycle.1 2 3 4 5 The advantage of segmental analysis of Color Kinesis images over acoustic quantification is that it allows quantification of the timing of endocardial motion during both contraction and relaxation on a regional basis.35 36 37
The morphology of fractional area change time histograms (Fig 5⇑) closely resembled that of the time derivatives of signal-averaged acoustic-quantification LV area waveforms38 and data obtained with radionuclide ventriculography.39 Integrating data with respect to time resulted in waveforms identical in shape to idealized LV area versus time plots (Fig 7⇑), including the rapid filling, diastasis, and atrial filling phases of diastole. However, in contrast to the above techniques, analysis of Color Kinesis images allows one to obtain quantitative information on the timing of regional LV function. Regional timing of ejection and filling was quantified by use of two parameters: (1) mean time (Fig 6⇑) and (2) time required to complete any specified fraction of area change, ie, time to 5%, 10%, 15%, …, 90%, 95%, and 100% of ejection or filling (Figs 7⇑ and 8⇑).
Pharmacological Interventions in Normal Subjects
Protocols 2 and 3 were designed to evaluate the ability of Color Kinesis to detect variations in spatial and, in particular, temporal patterns of regional endocardial motion induced by different pharmacological agents with known effects. Accordingly, we analyzed end-systolic and end-diastolic Color Kinesis images obtained in 14 normal subjects under esmolol and dobutamine to determine whether quantitative analysis of Color Kinesis images can objectively detect variations in LV systolic and diastolic performance induced by these agents. We found that in both systole and diastole, Color Kinesis identified the expected effects of dobutamine and esmolol on the magnitude and timing of endocardial motion. Our results obtained with atropine provided evidence that the above dobutamine-induced variations in the temporal characteristics of endocardial motion were not secondary to the chronotropic effects of the drug.
Although protocols 2 and 3 were aimed at studying the feasibility of quantitative segmental analysis of Color Kinesis images to track variations in regional LV performance, the pharmacological interventions used affect LV function globally. To evaluate the ability of this technique to reflect regional wall motion abnormalities related to regional myocardial ischemia37 40 41 42 rather than global variations on a regional basis, we applied this analysis to images obtained in patients undergoing routine dobutamine stress testing.
The results of protocol 4 demonstrated that systolic Color Kinesis images can be obtained under conditions of dobutamine stress testing. Segmental analysis of these images allowed objective detection of dobutamine-induced regional wall motion abnormalities in agreement with conventional visual interpretation of the corresponding two-dimensional images. Segments in which the two techniques disagreed were reexamined to find the reasons for discordance. We found that the reason for apparent false negatives was the discrepancy between segment size defined during visual interpretation of the two-dimensional echocardiograms and the segmentations used by the automated analysis. The apparent false-positive detections were a result of inadequate endocardial tracking due to suboptimal visualization of the endocardial boundary.
Our observations related to the timing of regional endocardial motion under conditions of dobutamine stress testing (Fig 14⇑) indicate that the effects of ischemia on the temporal sequence of systolic contraction may be biphasic: mildly hypokinetic segments show delayed motion, whereas the residual motion in severely hypokinetic or nearly akinetic segments occurred early in systole as assessed with this technique. To determine the clinical value of the temporal indices of systolic endocardial motion such as the regional mean time of ejection, these initial findings need to be confirmed in a larger population.
Although the sensitivity and specificity of this methodology cannot be determined from the results of this study, it is clear that this new technique is easy and objective and therefore may prove to be a useful clinical adjunct to conventional visual interpretation of two-dimensional images acquired during dobutamine stress testing.
Color Kinesis was designed to assess endocardial motion rather than wall thickening. The impact of this limitation on the diagnostic utility of this technique has yet to be determined. Moreover, like other echocardiographic techniques, the success of Color Kinesis is dependent on the quality of two-dimensional echocardiographic images.7 43 However, in our experience, it is possible to obtain Color Kinesis data in the majority of patients referred to our laboratory.
The low temporal resolution of Color Kinesis (30 frames per second) may not allow the definition of endocardial motion at high heart rates accurately enough to obtain additional indices of regional LV function, such as velocity of endocardial motion. On the other hand, the limited number of colors currently available for encoding may confound accurate acquisition of diastolic data in patients with low heart rates (<55 bpm), because the duration of color encoding may not be sufficient to cover the entire diastolic filling period. These limitations could be overcome by use of higher-frame-rate imaging in conjunction with an extended color scale.
Like other methods used to quantify endocardial excursion, Color Kinesis is affected by cardiac translation and/or rotation.19 20 Therefore, under conditions of stress echocardiography, when significant translation or rotation may be present, Color Kinesis may depict endocardial motion incorrectly. However, the results of our clinical trials confirm that Color Kinesis imaging under conditions of dobutamine stress is feasible and allows objective identification of 95% of regional wall motion abnormalities.
Respiration motion can also result in translation artifacts. These artifacts can be minimized by acquiring data during end expiration. Also, repeated acquisition strategy with data averaging, as used in this study, may ensure more reliable data acquisition. Another limitation of our method is the need for manual determination of anatomic landmarks used for segmentation. However, in our experience, these landmarks are usually easy to identify even in patients with suboptimal image quality.
To the best of our knowledge, this study is the first to develop and describe detailed techniques for the quantitative segmental analysis of magnitude and timing of endocardial motion based on Color Kinesis images. These techniques were initially evaluated by use of data obtained from small groups of normal subjects with good-quality echocardiographic images. Therefore, the results of this study should not be seen as an attempt to establish normal values for the different parameters. The number of subjects studied is not large enough to establish true confidence intervals for a normal population, which would require acquisition and analysis of data from large numbers of subjects over a wide age range. Similarly, a larger patient population would be required to accurately determine the sensitivity and specificity of this new methodology for objective interpretation of dobutamine stress echocardiography.
Future Applications of Color Kinesis
We believe that quantitative segmental analyses of Color Kinesis images is a promising clinical tool to quantify regional LV function. Analyses of magnitude of endocardial motion based on comparisons with normal values obtained in a large normal population may allow automated detection of regional wall motion abnormalities.16 Quantitative phase analysis of diastolic function, such as described in this study, in conjunction with high-frame-rate imaging,44 may allow objective, noninvasive assessment of regional relaxation abnormalities, in particular, during dobutamine stress testing. This is of particular clinical importance because ischemic heart disease first manifests itself as a regional diastolic dysfunction.35 45 Also, high-frame-rate imaging during systole could be useful for the assessment of conduction abnormalities. Acquisition and analysis of Color Kinesis images during dobutamine stress testing may improve the sensitivity and specificity of the diagnosis of coronary artery disease and facilitate the assessment of myocardial viability.46
In summary, Color Kinesis provides the basis for objective evaluation of regional endocardial wall motion. Detailed quantitative segmental analysis of Color Kinesis images promises to result in a variety of readily accessible indices of both magnitude and timing of endocardial motion that could be useful in a variety of pathological states. Future studies should be designed to explore the full clinical potential of this novel methodology.
The following appendix describes in detail how different parameters of magnitude and timing of regional as well as global endocardial motion were obtained from end-systolic and end-diastolic Color Kinesis images.
Incremental Fractional Area Change
In each segment, for each end-systolic or end-diastolic frame, the number of pixels of each color represents the incremental area change that occurred during a specific 33-ms period corresponding to that specific color. The end-diastolic area of each individual segment is represented by the total pixel count, ie, all colored pixels and those marked as blood. Normalization of the incremental area change by the end-diastolic area of the corresponding segment results in a regional fractional area change (in percent of end-diastolic area of that specific segment). Incremental area changes in all segments were displayed as a stacked color histogram, in which each 33-ms period is represented by a specific color identical to that used in Color Kinesis images. A histogram was also generated for fractional area change in percent of global rather than segmental end-diastolic area. For this purpose, global end-diastolic area was calculated as the sum of segmental end-diastolic areas in all six segments obtained as described above. The total area of this histogram reflects the traditional global fractional area change.
Incremental Endocardial Displacement
To evaluate regional endocardial radial displacement for each 33-ms period, arc approximation was used. The displacement, Δr, was calculated as the radial dimension of a respective color band according to the following expression: where rlong and rshort are the radial distances from the centroid, Along and Ashort are the corresponding sector areas, and θ is the sector angle (in radians). This equation was derived from the expression for an area of a sector with radius r: To obtain the displacement in metric units, areas Along and Ashort needed to be expressed in cm2 rather than pixel counts. Thus, each area was computed as a product of the corresponding pixel count with single pixel area, based on the 5:4 horizontal to vertical aspect ratio (NTSC standard) and imaging depth D (in cm) per 350 pixels: This conversion from pixels to metric units was performed for each consecutive color band in each segment, resulting in approximate endocardial wall displacements during each consecutive 33-ms period. This parameter was also displayed as a stacked color histogram in which each color represents a specific time during systole or diastole.
Incremental Fractional Radial Shortening
In each segment, fractional radial shortening was calculated for each 33-ms period from Equation 2, based on the area of the corresponding color band. To evaluate shortening in percent of end-diastolic radial dimension, wall displacement was normalized by segmental end-diastolic radius and multiplied by 100. The results were displayed as a color-encoded stacked histogram.
Temporal Patterns of Endocardial Motion
To directly assess the temporal patterns of LV contraction and relaxation, parameters of endocardial motion were displayed as a function of time. To obtain ejection and filling rates, incremental fractional area change (in percent of global end-diastolic area) was normalized by the time interval corresponding to a single video frame (33 ms). Ejection and filling rates were plotted versus time, with data from different segments displayed as stacked time histograms. This display allowed easy identification of both the magnitude and timing of peak ejection and filling.
Mean Time of Ejection and Filling
Color Kinesis tracks endocardial motion on the basis of pixel transitions between blood and myocardial tissue. Each pixel transition detected by the system may occur at a different time during systole or diastole and is color encoded accordingly. Therefore, mean time required for one pixel to change its tissue/blood attribute may be calculated in each segment from the colored pixel counts as follows: where RFAC(t) is a function of time representing regional fractional area change, time t=0 represents beginning of ejection or filling, and T is total ejection or filling time. This parameter represents the mean time of LV ejection or filling in each segment.
Integrated Fractional Area Change
Time curves reflecting fractional area change for the entire left ventricle integrated with respect to time were constructed to allow quantification of global LV function. These curves provided a simple display of the temporal progression of ejection and filling phases as well as the total fractional area change at the completion of contraction or relaxation, respectively.
Normalized Regional Time Curves
To evaluate the temporal patterns of contraction and relaxation on a segmental basis, individual time curves reflecting regional area change and radial shortening integrated with respect to time and normalized to 100% were constructed for each segment. To allow intrasubject and intersubject comparisons, the confounding effects of heart rate on the duration of systole and diastole were eliminated by linear interpolation to obtain 20 values of each parameter in 5% increments of contraction or relaxation time. In this display, each curve reaches 100% ejection or filling at 100% systolic or diastolic time, respectively.
These analyses were implemented in a user-friendly software analysis package designed for a personal computer. Time required to analyze one cardiac cycle (contraction and relaxation) was ≈2 minutes. Computing was completed in a fraction of a second in a Pentium 100-MHz personal computer.
We gratefully acknowledge the advice and support of David Prater, John Davidson, and Susan Floer from the Imaging Division of Hewlett Packard Company. We also thank the sonographers at the Noninvasive Cardiac Imaging Laboratories, Joanne Sandelski, Claudia Korcarz, and James Bednarz, for their help with data acquisition.
- Received June 27, 1996.
- Revision received October 17, 1996.
- Accepted October 23, 1996.
- Copyright © 1997 by American Heart Association
Perez JE, Klein SC, Prater DM, Fraser CE, Cardona H, Waggoner AD, Holland MR, Miller JG, Sobel BE. Automated on-line quantification of left ventricular dimensions and function by echocardiography with backscatter imaging and lateral gain compensation. Am J Cardiol. 1992;70:1200-1205.
Chenzbraun A, Pinto FJ, Popylisen S, Schnittger I, Popp RL. Comparison of acoustic quantification and Doppler echocardiography in assessment of left ventricular variables. Br Heart J. 1993;70:448-456.
Bazett HC. An analysis of the time-relations of electrocardiograms. Heart. 1920;7:353-370.
Haendchen R, Wyatt H, Maurer G, Zwehl W, Bear M, Meerbaum S, Corday E. Quantitation of regional cardiac function by two-dimensional echocardiography, I: patterns of contraction in the normal left ventricle. Circulation. 1983;67:1234-1245.
Pandian N, Kieso R, Kerber R. Two-dimensional echocardiography in experimental coronary stenosis, II: relationship between systolic wall thinning and regional myocardial perfusion in severe coronary stenosis. Circulation. 1982;66:603-611.
Schnittger I, Fitzgerald PJ, Gordon EP, Alderman EL, Popp RL. Computerized quantitative analysis of left ventricular wall motion by two-dimensional echocardiography. Circulation. 1984;70:242-254.
Ginzton L, Conant R, Brizendine M, Thigpen T, Laks M. Quantitative analysis of segmental wall motion during maximal upright dynamic exercise: variability in normal adults. Circulation. 1986;73:268-275.
Lang R, Vignon P, Weinert L, Bednarz J, Korcarz C, Sandelski J, Koch R, Prater D, Mor-Avi V. Echocardiographic quantification of regional left ventricular wall motion using Color Kinesis. Circulation. 1996;93:1877-1885.
Force T, Bloomfield P, O’Boyle JE, Khuri S, Josa M, Parisi A. Quantitative two-dimensional echocardiographic analysis of regional wall motion in patients with perioperative myocardial infarction. Circulation. 1984;70:233-241.
Moynihan P, Parisi A, Feldman CL. Quantitative detection of regional left ventricular contraction abnormalities by two-dimensional echocardiography, I: analysis of methods. Circulation. 1981;63:752-760.
Gillam L, Hogan R, Foale R, Franklin T, Newell J, Guyer D, Weyman A. A comparison of quantitative echocardiographic methods for delineating infarct-induced abnormal wall motion. Circulation. 1984;70:113-122.
Nidorf SM, Weyman AE. Left ventricle, II: quantification of segmental dysfunction. In: Weyman AE, ed. Principles and Practice of Echocardiography. Philadelphia, Pa: Lea & Febiger; 1994:625-655.
Gorcsan J III, Gasior TA, Mandarino WA, Deneault LG, Hattler BG, Pinsky MR. On-line estimation of changes in left ventricular stroke volume by transesophageal echocardiographic automated border detection in patients undergoing coronary artery bypass grafting. Am J Cardiol. 1993;72:721-727.
Parisi AF, Moynihan PF, Folland ED, Feldman CL. Quantitative detection of regional left ventricular contraction abnormalities by two-dimensional echocardiography, II: accuracy in coronary artery disease. Circulation. 1981;4:761-767.
Nieminen M, Parisi A, O’Boyle J, Folland E, Khuri S, Kloner R. Serial evaluation of myocardial thickening and thinning in acute experimental infarction: identification and quantification using two-dimensional echocardiography. Circulation. 1982;66:174-180.
Rushmer R, Crystal D, Wagner C. The functional anatomy of ventricular contraction. Circ Res. 1953;1:162-170.
Rogers WJJ, Shapiro EP, Weiss JL, Buchalter MB, Rademakers FE, Weisfeldt ML, Zerhouni EA. Quantification of and correction for left ventricular systolic long-axis shortening by magnetic resonance tissue tagging and slice isolation. Circulation. 1991;84:721-731.
Popp R. ‘High time’ for noninvasive assessment of regional ventricular diastolic ischemic dysfunction. Circulation. 1995;91:552-553.
Klausner S, Blair T, Bulawa W, Jeppson G, Jensen R, Clayton P. Quantitative analysis of segmental wall motion throughout systole and diastole in the normal human left ventricle. Circulation. 1982;65:580-590.
Weyman AE, Franklin TD, Hogan RD, Gillam LD, Wiske PS, Newell J, Gibbons EF, Foale RA. Importance of temporal heterogeneity in assessing the contraction abnormalities associated with acute myocardial ischemia. Circulation. 1984;70:102-112.
Udelson J, Bonow R. Radionuclide angiographic evaluation of left ventricular diastolic function. In: Gaasch W, LeWinter M, eds. Left Ventricular Diastolic Function and Heart Failure. Philadelphia, Pa: Lea & Febiger; 1994:167-191.
Schwartz S, Pandian N, Cao QL, Vannan M, Zachistal J, Simonetti J, Caldeira M. Automated on-line assessment of spatial and temporal abnormalities in regional left ventricular wall motion using color-encoded acoustic quantification of radiofrequency echocardiographic signals: description of the method and early clinical application. J Am Coll Cardiol. 1995;special issue:288A. Abstract.
Kondo H, Masuyama T, Ishihara K, Mano T, Yamamoto K, Naito J, Nagano R, Kishimoto S, Tanouchi J, Hori M, Takeda H, Inoue M, Kamada T. Digital subtraction high-frame-rate echocardiography in detecting delayed onset of regional left ventricular relaxation in ischemic heart disease. Circulation. 1995;91:304-312.
Pai RG. Echocardiographic assessment of myocardial viability. J Am Coll Cardiol. 1995;11:1-5.