カテゴリ分類を行わない評定方法 (連続確信度法) を用いた ROC 解析の有用性

概要

論文の詳細を見る
In the past years, receiver operating characteristic (ROC) studies that has been used in medical imaging system have involved data collected on a discrete confidence-rating scale with five categories, though in principle ROC studies which are called method for continuously-distributed test results can be conducted with continuous scale. In this study, we investigate the advantage of this new rating method. Advantages were obtained from our experiments that used both discrete and continuous rating scale with same sample sets. From the results of ROC analysis using the CR mammography with simulated mamma phantom, the method for continuously-distributed test results performs identical estimation of area under the ROC curve (Az) compared with the discrete confidence-rating method, but performs good (1/3) standard deviation of Az between readers. When the range of Az lies 0.65-0.70 which are "difficult" performance of ROC experiment, we concluded that the method for continuously-distributed test results has the advantages of good statistical power and reliability of results compared with the discrete confidencerating method.
公益社団法人日本放射線技術学会の論文
1994-10-01