用ROC和LROC方法观察三维CT成像识别缺陷的性能

ROC and LROC analysis of 3D CT performance for lesion detection

将ROC引入CT检测系统来对缺陷识别的效果进行定量分析。传统的ROC方法由于只针对缺陷位置固定的情形,存在自身的局限性,而考虑了缺陷定位的LROC方法更符合实际。本文基于两种方法获取的实验数据进行研究,并结合LROC模型和ROC模型的物理意义进行比较分析。结果表明,经综合比较,LROC方法优于ROC方法,但LROC模型的假设前提在一些情形下并不成立,因此在实际应用中也存在自身的局限性。目前来看,将ROC实验和LROC实验结合使用,是提高诊断效果的最优方案。

ROC curve is introduced to evaluate X-ray CT systems and to assess the detection performance by quantitative analysis.There are some limitations with traditional ROC method,since it only focuses on signal known exactly（SKE） situation.LROC method,which considers the location of defect,is closer to the actual situation.Both kinds of results were got by using ROC and LROC methods,and were compared with the physic meaning of ROC and LROC model.In general,LROC method was better than ROC,and the assumption of LROC model did not hold in some cases.There were also some limitations for LROC in practice.In present,combining ROC and LROC method is the way to get a nice result.