基于迁移学习的癫痫发作预测方法

Epilepsy seizure prediction method based on transfer learning

由于脑电信号因人而异,传统的机器学习方法很难适用于每一个患者。为了解决上述问题,文中提出一种基于迁移学习的癫痫发作预测方法。以波士顿儿童医院癫痫脑电数据集作为实验数据,以6组不同频率区间的带通滤波作为预处理方法,用样本熵作为分类特征。使用迁移学习后的VGG19网络作为分类器来识别发作期的癫痫脑电信号。该方法的最长预测时间为41.30 min,平均预测时间为23.82 min,最高预测准确率为93%,平均预测准确率为86.4%。最低误报率为22%,平均误报率为34%。实验结果表明,该方法可很好地用于癫痫发作的预测。

Because EEG signals vary from person to person,the traditional machine learning method is difficult to apply to every patient.In order to solve the above problems,this paper proposes a method of epileptic seizure prediction based on transfer learning.The EEG data set of CHB was used as the experimental data,six groups of band⁃pass filters with different frequency intervals were used as the preprocessing method,and the sample entropy was used as the feature selection.VGG19 network after transfer learning is used as classifier to recognize epileptic EEG signals.The longest prediction time of this method is 41.30 min,and the average prediction time is 23.82 min.The highest prediction accuracy was 93%,and the average prediction accuracy was 86.4%,the lowest false positive rate was 22%and the average was 34%.The experimental results show that this method can be well used in the prediction of epileptic seizures.