利用人工神经网络自动检测癫痫样放电的研究

AUTOMATIC DETECTION OF EPILEPTIFORM DISCHARGES USING AN ARTIFICIAL NEURAL NETWORK

根据对人工神经网络 (ArtificialNeuralNetwork ,ANN)识别脑电图 (EEG)的网络输出峰值分布曲线的分析 ,建立了一套利用ANN自动检测癫痫样放电 (EpileptiformDischarges,ED)的算法 ,即由计算机自动挑选ED模式 ,提取特征参数对网络进行训练 ,计算阈值 ,然后用训练好的网络和阈值对EEG进行识别。通过对典型的三种类型的癫痫患者的ED进行全自动识别 ,其平均识别率为 90 .9% ,平均假阳性率为 15 .7%。

This paper presents an approach to assist doctors' diagnosis of epileptiform discharges(ED) in EEG. The approach achieved automatic detection of ED by using an artificial neural network (ANN). The ANN was trained by the ED patterns selected according to the distribution curve of the peak value of the ANN output, with the parameters characters extracted. Then the threshold was calculated. The experimental results of three patients' EEG showed that the average hit rate (HR) was 91% while the average false positive rate (FPR) was 15%.