基于概率判决极端学习机的癫痫发作预报研究

平均预报敏感度和误报率是癫痫发作预报中最为重要的两个指标,针对在提高平均预报敏感度的同时误报率往往也会增高的问题,提出一种基于概率判决极端学习机的癫痫发作预报方法。该方法在利用平均相位相干指数对脑电信号进行特征提取的基础上,采用概率判决极端学习机进行分类,得到定量的分类信息之后,通过确定分类阈值来维持平均预报敏感度与误报率之间的平衡,最后经平滑过滤得到发作预报结果。对21例难治性癫痫病患者的仿真实验表明,本方法的平均预报敏感度可达到80.4%,平均误报率可低至0.10 h-1,具有较好的预报性能;而且训练时间短,为临床的在线应用提供了有价值的参考。

Epileptic Seizure Prediction Based on Probabilistic Discriminative Extreme Leaning Machine

Sensitivity and false-positive rate are two of the most important indicators of epileptic seizure prediction.When the sensitivity is improved,the false-positive rate will increase at the same time.To solve this problem,an epileptic seizure prediction method based on probabilistic discriminative extreme leaning machine(PDELM) was proposed.The method utilized PDELM for classification after extracting features from EEG by using mean phase coherence(MPC).And the probability of each class could be obtained.Then the balance of the sensitivity and the false-positive rate was maintained by determining a threshold.At last,after smoothing by a filter,the prediction results was obtained.Simulations on the 21 intractable epilepsy patients showed that the proposed method not only has a superior prediction performance(the mean sensitivity can reach 80.4% and the mean false-positive rate was as low as 0.10 h-1),but also required a short training time,which provided a valuable reference for the clinical application of epileptic seizure prediction.