影像学诊断评价中的参数法ROC曲线分析

目的 探讨采用ROC曲线参数分析法对影像学分类诊断结果进行评价的价值,并介绍ROC曲线参数分析软件ROCKIT。 方法 2名医师分别对60幅肺部CT图像进行肺结节良恶性5级分类诊断。分别用ROCKIT软件和SPSS软件对他们的诊断结果进行参数法和非参数法ROC曲线分析。 结果 对2名医师的诊断结果利用ROCKIT进行参数法ROC分析时,ROC曲线下面积分别为0.940±0.039和0.785±0.075(Z=2.056, P=0.040),利用SPSS进行非参数估计时结果分别为0.913±0.042和0.771±0.075。通过ROCKIT软件可绘制光滑的拟合ROC曲线,SPSS软件可绘制不光滑的经验ROC曲线。 结论 当有序分类资料样本量适中时,参数估计一般均无偏倚,非参数估计的结果可能小于真实值;ROCKIT软件是双正态参数法ROC曲线分析的有力工具。

Parametric ROC curve analysis for the performance evaluation of radiological diagnosis

Objective To explore the parametric ROC curve analysis for evaluating the performance of radiological diagnosis, and to present the usage of ROCKIT software. Methods Each nodule on 60 lung CT images were classified into one of the five possible classifications by two radiologists respectively. ROC curves and areas under the curves were estimated by using binormal parametric method via ROCKIT and nonparametric method via SPSS, respectively. Results Areas under ROC curves of the 2 radiologists by using ROCKIT was estimated at 0.940±0.039 and 0.785±0.075 (Z=2.056, P=0.040), respectively, while areas under the curves estimated by using SPSS was 0.913±0.042 and 0.771±0.075, respectively. Smooth fitted ROC curves were drawn with ROCKIT, whereas the unsmoothed empirical ROC curves were drawn with SPSS. Conclusion Parametric estimations of the ROC curve are un-bias when the sample size of ordinal scale data is moderate, whereas the nonparametric estimations may underestimate the real values. ROCKIT software is a powerful tool for binormal parametric analysis of the ROC curve.