文章摘要
徐绪党,刘 标,杨 伟,王传兵,孙 晋,李天女.影像学诊断试验的双正态模型参数法ROC曲线分析[J].南京医科大学学报,2014,(2):230~233
影像学诊断试验的双正态模型参数法ROC曲线分析
ROC curve analysis of dual-normal parameters model for imaging diagnostic tests
  
DOI:10.7655/NYDXBNS20140224
中文关键词: 诊断试验  受试者操作特性曲线  ROCKI程序  双正态模型
英文关键词: diagnostic tests  receiver operating characteristic curve  ROCKIT program  binormal model
基金项目:国家自然科学基金青年基金资助(81202032)
作者单位
徐绪党 南京医科大学第一附属医院核医学科,江苏 南京 210029 
刘 标 南京医科大学第一附属医院核医学科,江苏 南京 210029 
杨 伟 南京医科大学第一附属医院核医学科,江苏 南京 210029 
王传兵 南京医科大学第一附属医院放射科,江苏 南京 210029 
孙 晋 南京医科大学第一附属医院核医学科,江苏 南京 210029 
李天女 南京医科大学第一附属医院核医学科,江苏 南京 210029 
摘要点击次数: 845
全文下载次数: 529
中文摘要:
      目的:运用受试者工作特征(receiver operating characteristic,ROC)曲线的双正态模型参数法分析影像学诊断试验资料?方法:选择影像学上常见的单个诊断试验的连续性资料和两个完全相关诊断试验的有序分类资料进行分析;前者选取49例胃肿瘤性病变患者,进行常规MRI及弥散加权成像(diffusion-weighted image,DWI),分析胃良性和恶性占位之间的表观扩散系数(apparent diffusion coefficient,ADC)值差异,并确定其最佳截断点;后者选取55例孤立性肺结节(solitary pulmonary nodule,SPN)的HRCT及PET/CT资料,比较两成像模式的诊断效能?结合ROC分析软件ROCKIT行ROC分析,参考的 “金标准”皆为病理结果,差异性检验的水准α = 0.05?结果:前者诊断试验获得ROC曲线下面积Area(Az) = 0.963 5,ROC曲线相关参数a =2.347 6,相关参数 b = 0.844 7,约登指数最大值所对应的最佳截断点为1.915;后者诊断实验HRCT与PET/CT两诊断模式所获得的ROC曲线下面积分别为0.8127和0.959 0,相关系数为0.827 9;两者ROC曲线下面积的差异性Z检验,获得的单尾P值为0.000 1?结论:运用双正态模型参数法结合ROC曲线分析专用软件ROCKIT可以对影像学诊断试验资料进行系统分析,在单个诊断试验连续性资料的最佳工作点的确定及有序分类资料的诊断效能比较上可以实现有效的ROC评价?
英文摘要:
      Objective:To amaylze imaging diagnostic test data using receiver operating characteristic curve (ROC) combined with dual-normal model parameter method. Methods:Continuous data of a single common imaging diagnostic test and ordered segment information of two completely diagnostic tests were selected for analysis. The former group consisted of 49 cases of gastric neoplastic lesions was examined by conventional MRI and diffusion-weighted imaging(DWI) to analyze the difference of apparent diffusion coefficient (ADC) between gastric benign and malignant placeholder and determine the best cut-off point;the latter group consisted of 55 cases of Solitary pulmonary nodules(SPN) with HRCT and PET/CT data. Two types of data were analyzed by ROC analysis software (ROCKIT). The "gold standard" of diagnostic data was pathological results of significance(α = 0.05). Results:Area under the ROC curve(Az) of the former diagnostic tests was 0.963 5;the parameters a was 2.347 6,and the relevant parameter b was 0.844 7;Youden index corresponding to the maximum cut-off point was 1.915. The area under the ROC curve obtained by the latter diagnostic experiments HRCT and PET/CT were 0.812 7 and 0.959 0,respectively. The correlation coefficient was 0.827 9. Z test of the area under the curves of both group got the one-tailed P value of 0.000 1. Conclusion:The dual-normal parameters model combined with ROC curve analysis software ROCKIT could systematically analyze the imaging diagnostic data. ROC evaluation can be achieved in a single diagnostic test for continuous data to determine the optimum operating point and ordered categorical data on the diagnostic performance comparison.
查看全文   查看/发表评论  下载PDF阅读器
关闭