人工神经网络技术诊断附件包块良恶性的可行性研究  

Evaluation of artificial neural networks models for predicting ovarian malignancy in patients with adnexal masses

作　　者：王黎明[1] 付庆诏[1] 孔北华[2] 冯天瑾[3] 刘洪波[3] 周晓彬[4] 

机构地区：[1]青岛大学医学院附属医院,山东青岛266003 [2]山东大学齐鲁医院 [3]中国海洋大学 [4]青岛大学医学院

出　　处：《山东医药》2007年第22期19-21,共3页Shandong Medical Journal

摘　　要：目的探索运用计算机智能人工神经网络技术建立诊断模型来判断附件包块良恶性的可行性。方法180例附件包块患者随机分为训练组和测试组。训练组结合统计学多因素分析方法筛选出的参数建立人工神经网络诊断模型;测试组通过ROC曲线分析比较与恶性风险指数模型诊断性能的高低。结果18项指标中单因素分析有14项指标对判断附件包块的性质有意义(P<0.05)。由统计学多因素分析选出的参数(年龄、血清CA125、包块内壁突起、腹水及血流平均速度)共5项作为神经网络模型输入层参数,建立模型。ROC曲线证实神经网络模型较恶性风险指数模型的诊断性能高(P<0.05)。结论人工神经网络诊断模型对附件包块良恶性判断有良好的诊断性能。[ Objective] To evaluate the performance of artificial neural networks（ANN） models for predicting ovarian malignancy in patients with adnexal masses. [ Method] The data of 180 patients with adnexal masses were randomly divided into training and testing groups. The training subsets were used to screen out significant single parameters and to build the ANN＇s model combined with statistic method. The testing subsets were used to estimate the performance of ANN＇s model and risk of malignancy model （ RMI＇s model） （ ROC curve）. [ Result ] Fourteen of eighteen parameters were significant to diagnose malignancy in adnexal masses based on single analysis （ P 〈 0.05）. Three-layer back-propagation network（ ANN＇s model） based on the same input variables selected by multivariate analysis（ women＇s ages, CA125, papillary projection, ascites and time-averaged mean velocity of blood） had a significantly higher diagnostic rate than did RMI＇s model （P 〈 0. 05）. [ Conclusion ] Artificial neural network used in prediction of malignancy of adnexal masses have better diagnostic capability. There is a need for investigating it.

关 键 词：人工神经网络 恶性风险指数 附件包块 

分 类 号：R737.3[医药卫生—肿瘤]