ROC Analysis of Visual Assessments Made in Gated Blood Pool Scans of Patients with Coronary Artery Disease

관상동맥질환에서 심장풀 스캔의 육안적 평가에 대한 ROC 분석

Visual assessment of regional wall motion abnormality (RWMA) by gated blood pol scan (GBPS) serves as an useful parameter in the diagnosis, functional evaluation, and follow up in various clinical settings, but are still subject to some inherent limitations. On important problem may be the interobserver as well as intraobsever variation that may well be present due to the subjective nature of the interpretations. This study was carried out to determine the reliability and reproducibility of visual assessments made in GBPSs, and to observe the degree to which the results would be influenced by observer variation. Fifty two patients with coronary heart disease had resting GBPS and contrast ventriculography within 4 days appall. Contrast ventriculography-showed normal wall motion in 6 patients and the remaining 46 had RWMA in one or more segments. The anterior and left anterolateral views of all 52 GBPSs were analyzed by three independent observers, who selected from 5 scales, their level of confidence that there was RWMA in that segment. Reciever operating characteristic (ROC) curves for each analysis was plotted and the area under the curve $(\theta)$ was used as a parameter representing each observer's performance in his interpretations. The findings of contrast ventriculographies were used as the standard for RWMA. The apical and inferoapical segments showed the best correlation with contrast ventriculography ($\theta=0.90-0.94$, 0.81-0.94, respectively), and the inferior wall showed the poorest correlation $(\theta=0.70-0.74)$. The interpretations of the inferior, septal, apical, and posteroinferior, segments showed no difference between the observers, but there was significantly better performance in assessment by observer A compared to that by B or C for the anterolateral segments ($\theta=0.87$, 0.78, 0.76, respectively. p<0.01 for A vs B, p<0.05 for A vs C), as well as when all segments were considered altogether ($\theta=0.88$, 0.83, 0.82, respectively. both p<0.05). This was also true for the infero-apical segment between A and C ($\theta=0.09$, 0.81, p<0.05). The intraobserver variation, however, did not appear significant, with only the inferior segment for observer B showing any significant difference when observer A and B repeated the analysis 10 days latter. There was no difference in assessing dyskinesia, with all observers showing a high performance ($\theta=0.98$, 0.87, 0.97, respectively). The visual assessment of left ventricular ejection fraction by all three observers correlated well with the calculated value from a semiautomated method (Spearman's r = 0.91, 0.83, 0.83. p<0.01, p<0.05, p < 0.05). The assessment of LV and RV size also correlated well between the three observers (Kendall's w = 0.80, 0.51, p<0.01 for both left and right ventricles). The above findings suggest that RWMA visually assessed by GBPS correlates well with that done by contrast ventriculography. And although the observer's experience or skill may influence the results in certain segments, visual analysis of GBPS may serve as a reliable and reproducible means for evaluating ventricular function.