Sources of subjective variability in the assessment of left ventricular regional wall motion from contrast ventriculograms

End-diastolic and end-systolic volume from the left ventricular angiogram: how accurate is visual frame selection? Comparison between visual and semi-automated comnputer-assisted analysis

BACKGROUND

End-diastolic (ED), end-systolic (ES) left ventricular (LV) volumes and LV ejection fraction (LVEF) are important parameters for clinical decision making in heart disease. In clinical practice the frames from cine-angiography with the largest and smallest opacified LV areas are visually selected and the endocardial borders traced as LVED and LVES contours, respectively. We compared the accuracy of this visual method using two frames with a semi-automated computer assisted frame-by-frame analysis of the complete opacified cardiac cycles.

METHODS AND RESULTS

In 17 patients a biplane LV cine-angiogram was obtained at 25 frames/s. Complete frame-by-frame analysis was performed using semi-automatic border detection software. Experienced independent observers visually selected and manually traced LVED and LVES in the so-called visually assessed two-frame method in a consensus meeting. LV volumes were calculated by the area-length method. Mean LVEDV, LVESV and LVEF were 133 +/- 57, 56 +/- 40 ml and 61 +/- 16%, respectively, for the visually assessed two-frame method, and 117 +/- 49, 53 +/- 33 ml and 60 +/- 13%, respectively, for the semi-automated computer assisted frame-by-frame method. LVEDV was significantly higher in the visually assessed two-frame method (p < 0.01). Linear regression analysis showed an excellent correlation between semi-automated computer-assisted frame-by-frame and the visually assessed two-frame LVEDV (y = 1.2x - 2.9; r2 = 0.98), LVESV (y = 1.2x - 8.2; r2 = 0.97) and good linear correlation for LVEF (p = 1.2x - 3.6; r2 = 0.82). Bland-Altman analysis showed respectively a bias of 16.4, 2.4 ml and 5.0% with overall wide limits of agreement (-6.6 and 39.4 ml; -16.6 and 21.4 ml; -9.0% and 19.1%).

CONCLUSION

Correlation is excellent when visually assessed LVED and LVES are compared with a semi-automated computer assisted frame-by-frame analysis. However, the visually assessed two-frame method tends to overestimate the volumes obtained by semi-automated computer-assisted frame-by-frame analysis, especially for LVEDV, indicating that visual selection will yield a higher LVEF, which may influence clinical decision making.

Accuracy of subjective and computer-assisted assessments of angiographic left ventricular regional wall motion

To evaluate the accuracy of angiographic methods for analysis of left ventricular regional wall motion, we measured the ability of a carefully performed subjective analysis (three independent observers) and that of three computer-assisted methods (centerline, radial, and area) to detect the presence of significant coronary artery disease. Normal ranges were established in 90 studies showing normal wall motion, and accuracy was tested in a second, consecutive series of 43 ventriculograms. The results show that the subjective method best separated those regions with from those without significant coronary disease. The subjective method also demonstrated sensitivity comparable to the centerline and radial methods. Among the computer-assisted methods, the area method was least sensitive but most specific for both anterior and inferior region coronary disease. The centerline and radial methods were highly sensitive, but less specific. When coronary disease was defined by electrocardiographic Q-waves, the area method had superior accuracy. Further analysis showed the following: 1) For the centerline and radial methods, long-axis reregistration of the end-systolic frame resulted in loss of sensitivity but increased specificity for anterior wall coronary disease, and little change in analysis of the inferior wall; 2) inclusion or exclusion of the apex had little effect on predictive accuracy for anterior wall coronary disease; 3) as expected, sensitivity and specificity results of all methods were dependent on the selection of a normal range cutoff value; but when performance was optimized, the subjective and area methods had a higher overall predictive accuracy than the centerline method; 4) It is likely that all three computer-assisted methods could be calibrated to give indications of degrees of hypokinesis, as their results correlated well with subjective observers' evaluations over the entire range of scores. These results should assist in selection of the optimal method for regional wall motion analysis in clinical and research applications. As currently applied, the area method is the most specific of the computer-assisted methods tested; the centerline and radial methods have highest sensitivity and therefore are most suited to detection of mild degrees of hypokinesis.