Automatic Global and Regional Phase Analysis from Gated Myocardial Perfusion SPECT Imaging: Application to the Characterization of Ventricular Contraction in Patients with Left Bundle Branch Block

Patients

The total study population consisted of a consecutive group of 158 patients without known valvular disease or cardiomyopathy who underwent clinically indicated exercise or vasodilator rest/stress MPS at Cedars-Sinai Medical Center. Patients either had a normal stress electrocardiogram and low pretest likelihood of CAD (<5%) on the basis of age, sex, presenting symptoms, and coronary risk factors (23) (n = 86) or had LBBB without prior myocardial infarction or sternotomy (n = 72). All patients also had normal myocardial perfusion (summed stress score, <4) and normal LV ejection fraction (≥45%). Additional exclusion criteria for both groups were atrial fibrillation or frequent premature ventricular contractions that might have hampered the quantitative assessment of asynchrony using gated MPS. The patient characteristics are shown in Table 1.

Image Acquisition and Processing

Each patient underwent a standard separate-acquisition dual-isotope rest/stress ²⁰¹Tl/^99mTc sestamibi protocol as previously described (24), using a dual-detector camera (Forte; Philips, or e.cam; Siemens Medical Systems), 90° configuration, elliptic 180° orbit, 64 projections, 25 s per projection, and 16 frames per R-R interval. For this study, only stress images were analyzed. For stress imaging, ^99mTc-sestamibi (925–1,295 MBq; dose adjusted to patient weight) was injected at peak exercise or at peak adenosine effect (2 min). Gated MPS acquisition was initiated 15–30 min after stress, using a 20% energy window centered over the 140-keV photopeak. The beat length acceptance window was set to 100%, such that beats ranging from 50% to 150% of the mean beat length were accepted. Projection images were reconstructed using filtered backprojection (ramp filter followed by 2-dimensional Butterworth filtering; order, 5; cutoff frequency, 0.5 cycle per pixel on a 0–1 scale; pixel size, 6.5 mm) and automatic oblique reorientation (25) to produce gated short-axis image datasets. Neither attenuation nor scatter correction was performed.

Phase Parameters

For this study, gated short-axis images were segmented using a modified version of QGS that provides new global and regional quantitative parameters for the assessment of intraventricular dyssynchrony as well as previously described functional parameters (26,27). QGS operates by first computing a midmyocardial LV surface. QGS then computes endocardial and epicardial surfaces using count-profile and thickening information derived from myocardial mass conservation combined with count increases caused by partial-volume effects. For the phase-analysis algorithm, a unidimensional array is created for each spatial sampling point that contains the local maximum myocardial counts at each interval, obtained by searching between the epicardium and endocardium along a direction normal to the midmyocardial surface. This array represents a time-varying, periodic function from which the first Fourier harmonic (FFH) is determined, the phase angle of which forms the basis of all of the synchrony measurements. Amplitude information is used by the algorithm to eliminate phase measurements for which the corresponding amplitude was deemed too low to yield accurate measurements: 5% of the samples corresponding to the lowest amplitudes across the whole myocardium are automatically removed from the analysis because phase measurements are essentially ill-defined for signals that exhibit low temporal variations (13).

Global and Regional Synchrony Parameters

Once timing parameters have been calculated for all sampling points, multiple global and regional LV synchrony measures are computed on the basis of whole-ventricle, wall-based, vessel-based, and segment-based (17 or 20 segments) statistics. These statistics are calculated for the histogram (a tabulation of distribution frequencies) of FFH phase angles, and their notations in this article are as follows: μ _R, mean over region R; σ _R, SD over region R; M_R, mode over region R (i.e., the location of the peak of the phase-angle histogram for region R); β_R, bandwidth of the phase-angle histogram over region R (a measure of variability described later in this article); and ε_R, entropy of the phase-angle histogram over region R (a measure of variability described later in this article).

If the region R is not specified for a parameter (e.g., μ instead of μ_R), the statistics are computed over the whole left ventricle for a maximum of 28 × 36 samples, depending on the geometry of the LV valve plane (22). The mean, SD, and mode are calculated in the standard manner. The bandwidth, β, of a phase histogram is defined as the smallest phase-angle range that encompasses 95% of the histogram samples (19). The entropy, ε, of any given variable is computed as previously described in the context of information theory (28). ε is normalized to its maximum value for the number of histogram bins used (arbitrarily fixed at 60) and reported as a percentage. This measure of variability has been described as potentially more sensitive than the SD of the phase (13), which has often been used in the assessment of cardiac synchrony (29,30) and is equal to zero if only 1 phase value is present in the histogram and positive otherwise. The formula used to compute ε is shown in Equation 1, in which n represents the number of histogram bins, f_i is the frequency of the histogram bin i (i.e., the number of measured phase angles that fall within bin i divided by the total number of phase-angle measurements in the histogram; range, 0…1), and log_e x is the natural logarithm of number x:Eq. 1

Phase-angle polar maps and global synchrony histograms for representative patients with LLk and LBBB are shown in Figure 1.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1. 

Comparison of global and regional thickening phase in individual patients from LLk and LBBB groups. Global histograms (top row) show overall increased dyssynchrony in patient with LBBB, and polar maps (middle) show uniform contraction for patient with LLk and clear septal-to-lateral contraction delay for patient with LBBB. Color scale used for both patients (bottom) has been adjusted to emphasize contraction timing differences in middle third of R-R cycle. deg = degrees.

In addition to whole-ventricle statistics, we included in this study differences between regional measures computed for the lateral wall (LAT) versus the septal wall (SEP) and differences between the average of segments 11 and 12 (midventricular lateral) versus segments 8 and 9 (midventricular septal) of a 17-segment model (Fig. 2). These differences were calculated for the means and modes of the chosen parameters and reported in this article as Δμ and ΔM, respectively, with the subscripts W for wall-based calculations and S for segment-based calculations: Δμ_W = μ_LAT − μ_SEP, ΔM_W = M_LAT − M_SEP, Δμ_S = (μ₁₁ + μ₁₂)/2 − (μ₈ + μ₉)/2, and ΔM_S = (M₁₁ + M₁₂)/2 − (M₈ + M₉)/2.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2. 

Wall and segment definitions used for regional calculations. In wall model, differences between lateral (LAT) and septal (SEP) walls are used; anterior (ANT), inferior (INF), and apical (APX) regions of LV are not used in this analysis. In American Heart Association 17-segment model, lateral measurements are averaged from segments 11 and 12, and septal measurements are averaged from segments 8 and 9. AHA = American Heart Association.

Timing values can be expressed in milliseconds, as angles in degrees (range, 0°−360°), or as a percentage of the average duration of the cardiac cycle (range, 0%–100%). Although timing values expressed in milliseconds may be clinically relevant (31), they are directly dependent on heart rate, potentially rendering comparison of studies acquired from patients with different heart rates problematic (10). Because the purpose of this study was not to quantify absolute timing delays but to compare differences between groups, we have expressed all timing values in degrees.

Statistical Analysis

Before comparing the LLk and LBBB groups, we performed intragroup comparisons between the sexes using 2-sample, unpaired t tests to determine whether the parameters analyzed were sex-specific and required separating patients into male and female subpopulations. Subsequently, the LLk and LBBB groups were compared by choosing contraction timing–related variables of interest and performing 2-sample, unpaired t tests. Welch's correction was applied if variances were significantly different. Receiver-operating-characteristic (ROC) analysis was also performed. LLk and LBBB denote the combined (male plus female) populations, and the M and F subscripts denote the male and female subpopulations, respectively. This yields the following groups: LLk, LBBB, LLk_M, LLk_F, LBBB_M, and LBBB_F. Calculations were performed using GraphPad Prism (version 5; GraphPad Software) and Stata (version 8; StataCorp LP).

Sex Differences

Table 2 summarizes the overall findings by sex within the LLk and LBBB groups. In the LLk group, the global parameters tended to be sex-specific, with statistically significant differences between men and women; differences between regional values were sex-independent, with no statistically significant differences for any of the regional parameters. In the patients with LBBB, the sex differences were minimal. Because of the findings in the LLk patients, we report below sex-specific results for global measures and combined results for regional differences.

Global Parameters

Table 3 summarizes t test and ROC analysis results for global intraventricular sex-specific dyssynchrony parameters. Significant differences were found for all parameters (σ, β, and ε) between LLk_F and LBBB_F (women) and between LLk_M and LBBB_M (men). Data points, interquartile ranges, and ROC curves are shown in Figure 3.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 3. 

Data points and ROC analysis for global parameters σ (top), β (middle), and ε (bottom). Actual measurements for patients in each group are shown on left, with lines indicating median and interquartile range for each cluster, and ROC curves for men and women are shown on right.

Regional Parameters

Table 4 summarizes t test and ROC analysis results for sex-independent differences between regional FFH phase-angle means and modes. Significant differences were found for all parameters (Δμ_W, ΔM_W, Δμ_S, and ΔM_S) between LLk and LBBB. Data points, interquartile ranges, and ROC curves are shown in Figure 4.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 4. 

Data points and ROC analysis for regional parameters. Wall-based measurements (Δμ_W and ΔM_W) are shown on top, and segment-based measurements (Δμ_S and ΔM_S) are shown on bottom. Actual measurements for differences of means and modes are shown on left, with lines indicating median and interquartile range for each cluster, and ROC curves are shown on right.

Global Parameters

In both the female and the male LLk groups, t tests showed that P values for ε were lower than those for σ and β, and ROC analysis demonstrated increased area under the curve (AUC), sensitivity (Ss), and specificity (Sp) for ε versus σ and β. These results can be explained by examining how σ, β, and ε are affected by the distribution of phase histogram values. ε is a mathematical expression of synchrony that does not rely on phase similarity. The contribution to ε of 2 contiguous histogram bins is the same as the contribution to ε of 2 distant histogram bins; therefore, ε is not influenced by the presence of outliers (histogram bars located farther from the histogram peak), whereas these bins increase the value of σ and β to a greater extent. It has also been reported that such a measure better discriminates between multiple pathologies, not solely between normal and abnormal contraction patterns (13).

Regional Parameters

Differences between lateral and septal regional measurements were analyzed for combined female plus male populations. All differences were statistically significant (P < 0.0001), and ROC AUCs were all greater than 0.92, with high Ss (>87%) and Sp (>85%) for all parameters. Overall, mode differences yielded a slightly higher AUC than did mean differences. Segment-based calculations also yielded a marginally higher AUC than did wall-based calculations but not necessarily higher Ss and Sp. One might expect slightly better results with segment-based calculations because the chosen midventricular segments exclude the more troublesome basal region of the LV in which segmented contours, and therefore extracted counts, may be less accurate as the myocardium becomes thinner. A comparison of ROC curves, however, indicated that differences were not statistically significant.

Average thickening timing differences were close to zero for patients with LLk (−1.1°, 0.7°, −1.7°, and −2.3° for Δμ_W, ΔM_W, Δμ_S, and ΔM_S, respectively) and positive for patients with LBBB (32.7°, 32.3°, 29.2°, and 29.2° for Δμ_W, ΔM_W, Δμ_S, and ΔM_S, respectively). As previously mentioned, positive differences indicating later contraction in the lateral wall than in the septal wall were to be expected given the physiopathology of LBBB. Figure 5 shows a sample comparison of wall-based thickening (counts) curves for individual patients with LLk and LBBB, graphically emphasizing regional differences between the timing of thickening for the LV walls.

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 5. 

Comparison of regional count curves vs. time (expressed in degrees, from 0° to 360°) in individual patients from LLk and LBBB groups. Vertical count scale is omitted for clarity. Inner area of each outline (A and B) represents 5° spread and is used to further emphasize delay in propagation of mechanical contraction wave from septum to inferior and anterior walls, and lastly to lateral wall for LBBB patient. ANT = anterior; INF = inferior; LAT = lateral; SEP = septal.

Study Limitations

Myocardial stunning in the poststress images was not a concern as it might be in a general population of patients with CAD because perfusion and function were normal for all patients, and visual transient ischemic dilation, an indicator of possible balanced reduction of flow, was not present. In other studies, and especially when comparing SPECT dyssynchrony parameters with other measurements acquired at rest, it may be advisable to process rest images instead of poststress images.

Because our algorithm relies on myocardial count variations, severe and extensive ischemia can be detrimental to the quantification of dyssynchrony. Although the algorithm handles hypoperfused areas by eliminating a fixed proportion of measurements instead of relying on a threshold that might discard large portions of the myocardium, in extreme cases it may be necessary to rely on another technique to derive dyssynchrony measurements. Preliminary experience with patients with chronic heart failure, however, indicates that this may only rarely be necessary.

The FFH phase angle derived from wall thickening information is often used to quantify the onset of ventricular contraction (19). This angle, however, can be influenced by a variety of other timing factors (33), a problem that can be more easily alleviated in high–frame-rate imaging techniques such as radionuclide angiography by discarding a fraction of the late diastolic filling phase frames (15).

As has been previously described (27), accurately measuring thickness from MPS is intrinsically difficult. This study, however, is concerned only with variations of this quantity, and it is reasonable to expect that the temporal behavior of thickness curves, approximated by myocardial counts, adequately depicts the underlying physiologic process of myocardial contraction. At the same time, it is also reasonable to expect that myocardial wall motion may be affected by other phenomena such as paradoxic stretching of the early activated territory (14) in the case of LBBB, and therefore wall motion was not included in this analysis. A discussion of normal regional motion in QGS is available elsewhere (34).

Differences between regional SD, bandwidths, or entropies were not included because they would not have a meaningful interpretation in this context. Similarly, global LV means and modes are not reported, because they do not provide information regarding the degree of dyssynchrony.

MPS currently provides only LV synchrony assessment, but a prognostic study using radionuclide cineangiography for an idiopathic dilated cardiomyopathy population showed intraventricular synchrony to be a better predictor than interventricular synchrony (35), indicating that the lack of interventricular measurements may not be a relevant limitation.