Intra- and inter-operator repeatability of myocardial blood flow and myocardial flow reserve measurements using rubidium-82 pet and a highly automated analysis program

Abstract

Background

Changes in myocardial blood flow between rest and stress states are commonly used to diagnose coronary artery disease. Relative myocardial perfusion imaging (MPI) is used routinely while myocardial blood flow quantification (MBF) may improve the sensitivity for detection of early disease. The ratio of flow at stress and rest (S/R) and their difference (S-R) have both been proposed as a means to detect regions with reduced myocardial flow reserve (MFR). In this study, we describe a highly automated method to calculate regional and global rest, stress, S/R, and S-R polar maps of the left ventricle myocardium.

Methods

We measured the inter- and intra-operator variability using two randomized datasets (n = 30 each) for each of two operators (novice and expert) with correlation and Bland-Altman reproducibility coefficient (RPC%) analyses.

Results

S-R MBF had less inter-operator dependent variability than S/R (RPC% = 5.0% vs 12.6%, P < .001). While there was no difference in intra-operator variability with S-R MBF (novice vs expert RPC% = 6.4% vs 5.9%, P = ns), variability was higher in the novice-operator for S/R (RPC% = 16.8% vs 8.5% respectively, P < .001), suggesting that S-R may be preferred for detecting small changes in MFR. The novice operator’s intervention pattern became more similar to that of the expert in the later dataset, emphasizing the need for adequate training and quality assurance.

Conclusion

The proposed method results in low operator-dependent variability, suitable for routine use.

Appendix

The spline optimization algorithm minimized a cost function, C _energy, that resulted in maximization of the image energy overlapping the spline model. Penalties were applied to discourage abnormal myocardial shapes by minimizing the following metrics:

• 1. Eccentricity of SA: the LV should be somewhat circular, thus if slices with a variation of radii greater than 30% exist, a penalty was applied.

  $$ C_{\text{elip}} = \left\{ {\begin{array}{ll} 0 & {e < 0.3} \\ e & {e \ge 0.3} \\ \end{array} } \right.\quad e = \text{max}_{i} \left[ {\left| {\log (r_{{{\text{hor}}_{i} }} /r_{{{\text{ver}}_{i} }} )} \right|} \right] $$

  (6)

• 2. Relative size of atrium: the cross section of the atrium should not be bigger than that of the ventricle, thus a penalty was applied if the mean of its radii was more than 20% larger than the mean of the radii of the basal and cavity sections.

  $$ C_{\text{atrium}} = \left\{ {\begin{array}{ll} 0 & {a < 1.2} \\ a & {a \ge 1.2} \\ \end{array} } \right.\quad a = \frac{1}{2}{\frac{{r_{{{\text{hor}}_{\text{atrium}} }} + r_{{{\text{ver}}_{\text{atrium}} }} }}{{r_{{{\text{hor}}_{\text{cavity}} }} + r_{{{\text{ver}}_{\text{cavity}} }} + r_{{{\text{hor}}_{\text{base}} }} + r_{{{\text{ver}}_{\text{base}} }} }}} $$

  (7)

• 3. Offset of center of ellipse from LV long axis: the LV myocardium should be nearly centered on the LV long axis, thus a penalty was applied if the center of the myocardium was displaced from the LV long axis by more than 40% of the mean radius in the same slice.

  $$ C_{\text{offset}} = \left\{ {\begin{array}{ll} 0 & {o < 0.4} \\ o & {o \ge 0.4} \\ \end{array} } \right.\quad o = \text{max}_{i} \left[ {{\frac{{2o_{i} }}{{r_{{{\text{ver}}_{i} }} + r_{{{\text{hor}}_{i} }} }}}} \right] $$

  (8)

The final cost function, C, defined by Eq. 9 accounted for all the above penalties while rewarding for energy overlapping the LV model. Thus, the LV model was constrained to have a characteristic shape, but abnormal myocardial shapes could be accommodated by the model, provided the image intensity is sufficient to offset the penalties.

$$ C = C_{\text{energy}} \left( { 1+ 10 \times C_{\text{elip}} + C_{\text{atrium}} + C_{\text{offset}} } \right) $$

(9)