[Variances among readers analyzed using the Jackknife method in ROC analysis].

To improve the reliability of statistically significant test results between two imaging systems in receiver operating characteristic (ROC) analysis, it was important to evaluate the variances among readers. In this study, we tested statistically significant differences with actual experimental data from ROC studies that were performed to evaluate the detectability of 4 kinds of image intensifier fluorographic films. The DBM MRMC--i.e., the software for analyze of statistically significant differences by the Jackknife method--was used. The relationship between p values obtained from analysis that treated only readers as a random sample in DBM MRMC, and the variances among readers, were investigated. The testing results showed that, when there were slight differences in the mean area under the corve (AUC) between imaging systems, variances among readers were estimated by p values, because p values indicated the degree of variance among readers. Furthermore, we found that p values obtained from analysis that treated only readers as a random sample in DBM MRMC were equal to p values of the two-tailed paired t-test. It was considered that variances among readers could be analyzed by investigating the relationship between the differences in the mean AUC of two imaging systems and the p values obtained in a two-tailed paired t-test.