摘要
目的基于分类与回归决策树(CART)与人工神经网络(ANN)技术建立孤立性肺结节良恶性的诊断模型,探讨数据挖掘技术在孤立性肺结节影像学诊断中的应用。方法收集经病理学证实的58例孤立性肺结节(SPN)患者资料,分别提取12个临床指标和22个影像学指标作为鉴别SPN良恶性的输入指标,将各指标结果输入CART和ANN诊断模型。采用ROCKIT统计学软件绘制三组影像科医生、CART和ANN的受试者操作特征(ROC)曲线。结果CART对SPN良恶性诊断正确率最高,其次为ANN、高年资医生、中等年资医生和低年资医生。以上各组ROC曲线下面积分别为0.931、0.878、0.845、0.778和0.658。CART、ANN与高年资医生相比无显著性差异(P>0.05);但三组与中、低年资组医生相比有显著性差异(P<0.05);CART对SPN具有决策意义的诊断指标为年龄,其次为结节的毛刺征和咯血症状。结论数据挖掘的CART和ANN两种算法对评估孤立性肺结节的良恶性具有较高的准确性。
Objective To establish and compare classification and regression tree （CART） and artificial neural network （ANN） and in differentiating benign from malignant solitary pulmonary nudules （SPN） on CT findings, thus investigating the application of data mining techniques in the application of imaging diagnosis. Methods Fifty-eight cases with pulmonary nodules proved by pathology were studied. A CART and ANN were established to distinguish benign from malignant SPN on the basis of 12 clinical parameters and 22 radiological findings that were extracted by chief radiologists according to standard references. The receiver operating characteristic （ROC） curve analysis of radiologists, CART and ANN were analyzed. Results CART and ANN showed a high performance in differentiating benign from malignant pulmonary nodules （Az= 0. 931 for CART, Az=0. 878 for ANN, P〉0. 05）. They showed no significant difference between chief radiologists （Az= 0. 845, P〉0.05）, but the performance of ANN, CART as well as the chief radiologists was higher than that of attending radiologists （Az=0. 778, P〉0.05） and radiology residents （Az=0. 658, P〈0.05）. By means of CART, the most important decision-making variable is age, next is spiculation and hemoptysis. Conclusion Data mining techniques using CART and ANN prove a high accuracy in differentiating benign from malignant pulmonary nodules based on clinical variables and CT findings.
出处
《中国医学影像技术》
CSCD
北大核心
2008年第3期438-442,共5页
Chinese Journal of Medical Imaging Technology
基金
国家自然科学基金项目(30470509)
关键词
体层摄影术
X线计算机
孤立性肺结节
分类与回归树
人工神经网络
数据挖掘
Tomography, X-ray computed
Solitary pulmomary nodules
Classification and regression tree
Artificial neural network
Data mining
