摘要
目的评价人工神经网络模型在超声诊断早期肝硬化的准确性和可靠性。方法收集临床慢性肝病患者超声影像特征数据为训练样本集,以肝脏组织病理学活检作为金标准,建立超声诊断早期肝硬化人工神经网络模型;同时,将另外收集的慢性肝病患者超声影像特征数据通过神经网络进行仿真运算,得到网络诊断输出,评价该神经网络模型应用于早期肝硬化超声影像诊断的准确性和可靠性。结果以统计学选出的参数作为输入层参数的三层人工神经网络用于超声诊断早期肝硬化的灵敏度和特异度分别为84.1%、84.6%,准确度87.9%,尤登指数0.687。结论人工神经网络模型用于超声诊断早期肝硬化具有较高的准确性,有可能为肝硬化早期诊断提供又一实用有效的手段。
Objective To evaluate the validity and reliability of ultrasonic diagnosis of earlier period hepatic cirrhosis model on artificial neural network（ANN）. Methods The ultrasonic features of all the patients with chronic liver diseases were collected for the training database, pathologic biopsy of liver tissue was used as gold standard. The ANN for ultrasonic diagnosis of earlier period hepatic cirrhosis was constructed with a back- propagation training algorithm. Based on the emulational operation by ANN, a network diagnosis was made. The validity and reliability of ANN for earlier period hepatic cirrhosis ultrasonic diagnosis was calculated. Results Three- layer back- propagation network was based on the input variables selected by using statistics analysis, the sensitivity, specificity and accuracy for ultrasonic diagnosis of earlier period hepatic cirrhosis were 84.1%, 84.6% and 87.9%, Youden index was 0. 687. Conclusion It has high accuracy in diagnosis of earlier period hepatic cirrhosis, and it is an efficient means for earlier diagnosis of hepatic cirrhosis.
出处
《临床超声医学杂志》
2008年第8期517-519,共3页
Journal of Clinical Ultrasound in Medicine
基金
陕西省科技计划项目[2006k12-G2(2)]
关键词
人工神经网络
肝硬化
早期
超声检查
Artificial neural network
Hepatic cirrhosis,earlier
Ultrasounography
