论文部分内容阅读
目的：研究基于神经网络的视觉诱发电位波形的自动识别方法，实现Ｐ１００波的自动识别。方法：从历年来临床闪光ＶＥＰ检查记录中选择出经专家用视觉方法判断有Ｐ１００波存在的记录文件，记录专家判别的Ｐ１００波和比较波的数据，以此记录材料作为神经网络的学习样本，建立采用反向回归算法的神经网络。选取波形特征参数作为神经网络的输入参数，用具有不同隐层结点数的神经网络对相同样本进行学习和检验。结果：神经网络对学习样本的判别一致率为：Ａ组９８８％，Ｂ组９８８％；对检验样本的判别一致率为：Ａ组平均９２２％，Ｂ组平均９６０％。以全体材料为学习样本时网络的判别一致率为９７６％～９８２％，其中不同隐结点数的网络判别一致率差异极显著（χ２检验，χ２＝９．６３＞６．６３，Ｐ＜０．０１）。结论：专家的判断原则可以为神经网络所阐明
Objective: To study the automatic identification method of visual evoked potential waveform based on neural network to realize the automatic identification of P100 wave. Methods: From the records of clinical flash VEP over the years, the experts selected the records of P100 wave by visual method and recorded the data of P100 wave and comparative wave determined by experts. Using this recorded material as the learning sample of neural network, Establish a neural network using inverse regression algorithm. The waveform characteristic parameters are selected as the input parameters of the neural network, and the neural network with different hidden layer nodes is used to study and test the same samples. Results: The discriminant concordance rate of neural network to learning samples was 98.8% in group A and 98.8% in group B, and the discriminant agreement rate of test samples was 922% in group A and 960 in group B %. The discriminant concordance rate of the whole network was 97.6% ~ 982% when all the samples were used as learning samples. The difference of the discriminant agreement rate of network with different hidden nodes was significant (χ2 test, χ2 = 9.63> 6.63, P <0.01). Conclusion: Expert judgment principle can be clarified by neural network