【摘要】 目的:综合运用模糊数学和人工神经网络知识构建一个模糊神经网络(FNN)模型,用于肺癌计算机辅助诊断。方法:以实际肺癌诊断病例(n=117)中的一部分(n=73)作为训练集,首先利用隶属度函数对样本的5个临床参数和21项CT特征进行模糊化处理,再输入基于BP算法的神经网络,对网络进行训练。用训练好的网络对余下的样本(n=44)进行预测,并将预测结果以及基于BP神经网络(BPNN)的预测结果与病理结果进行比较。结果:FNN诊断肺癌的灵敏度、特异度和正确率分别为0.904 8、0.913 0和90.91%,BPNN分别为0.809 5、0.869 6和84.09%。结论:FNN模型诊断肺癌的预测结果与病理结果接近,且优于BPNN的预测结果。更多还原

【Abstract】 Aim: To develop a computer-aided scheme of the lung cancer diagnosis by CT based on fuzzy neural net-works( FNN) to assist radiologists in distinguishing malignant tumor from benign pulmonary nodules. Methods: With a part of actual lung cancer diagnosis cases as samples( n = 73),first,sample data were treated with membership function.Then,the treated data were used as input of the neural networks based on back-propagation algorithm and the neural networks was trained. The other cases( n = 44) were used as validation data and were forecasted by the trained FNN. The result of FNN and that of the usual back-propagation neural network( BPNN) were both compared with the pathological results. Results: The sensitivity,specificity and accuracy of the FNN in the diagnosis of lung cancer were 0. 904 8,0. 913 0and 90. 91%,and those for BPNN were 0. 809 5,0. 869 6 and 84. 09%. Conclusion: The forecasting results by FNN is more consistent with that of pathology than that by the BPNN model.更多还原