Kors J A, van Herpen G, Sittig A C, van Bemmel J H
Department of Medical Informatics, Faculty of Medicine and Health Sciences, Erasmus University, Rotterdam, The Netherlands.
Eur Heart J. 1990 Dec;11(12):1083-92. doi: 10.1093/oxfordjournals.eurheartj.a059647.
Three methods for reconstructing the Frank VCG from the standard 12-lead ECG were studied. The first was based on multivariate regression, the second on a model of the cardio-electrical activity, and the third method used a quasi-orthogonal set of ECG leads. The methods were evaluated on a test set of 90 cases by a numerical distance measure and by the agreement in diagnostic classification of the original and reconstructed VCGs. The original and reconstructed VCGs were presented separately and in random order to three referees. Eighteen of the original VCGs were presented three times to estimate the intra-observer agreement. Kappa statistics were used to quantify the agreement between diagnostic classifications. Separately, one referee was simultaneously presented the original VCG and its three reconstructions for all cases. Each reconstruction VCG was classified as either diagnostically 'same' as the original, 'borderline' or 'different'. The performance of the regression method and the model-based method was comparable. Both methods were preferable to the quasi-orthogonal method. The kappa values for the preferred methods indicated a good to excellent diagnostic agreement between the original and reconstructed VCGs. Only one out of ninety VCGs that were reconstructed with the regression method was classified as 'different' compared with the original VCGs; three VCGs were classified as 'different' with the model-based method. It was also found that estimation of similarity by a distance measure could not replace diagnostic evaluation by skilled observers.
研究了三种从标准12导联心电图重建Frank心向量图（VCG）的方法。第一种基于多元回归，第二种基于心脏电活动模型，第三种方法使用了一组准正交的心电图导联。通过数值距离测量以及原始VCG和重建VCG在诊断分类上的一致性，在90例测试集上对这些方法进行了评估。原始VCG和重建VCG以随机顺序分别呈现给三位裁判。18个原始VCG被呈现了三次，以估计观察者内部的一致性。kappa统计量用于量化诊断分类之间的一致性。另外，一位裁判同时看到了所有病例的原始VCG及其三种重建结果。每个重建的VCG被分类为在诊断上与原始VCG“相同”、“临界”或“不同”。回归方法和基于模型的方法的性能相当。这两种方法都优于准正交方法。首选方法的kappa值表明原始VCG和重建VCG之间在诊断上具有良好到极好的一致性。与原始VCG相比，用回归方法重建的90个VCG中只有1个被分类为“不同”；用基于模型的方法有3个VCG被分类为“不同”。还发现，通过距离测量来估计相似性不能替代由熟练观察者进行的诊断评估。
