基于SAE深度神经网络的自主循环恢复辨识算法研究

Research on identification algorithm of return of spontaneous circulation based on stacked autoencoder deep neural network

目的:提出一种新的胸部按压过程中人体自主循环恢复辨识算法,以实现对按压急救过程中自主循环恢复的准确辨识,指导胸部按压的启停。方法:对10头实验猪进行室颤诱导及电击除颤以采集包含自主循环恢复阶段的数据,期间持续记录脉搏波信号的变化,数据采集完成后对脉搏波进行处理并得到能量图谱,基于栈式自编码(stacked autoencoder,SAE)深度神经网络进行建模预测,实现自主循环恢复的准确辨识。结果:实验结果显示,在心脏骤停心肺复苏自主循环恢复过程中,基于SAE深度神经网络的自主循环恢复辨识算法对无胸部按压情况下自主循环恢复的平均辨识准确度达95.0%,对有胸部按压情况下自主循环恢复的平均辨识准确度达86.5%。结论:基于SAE深度神经网络的自主循环恢复辨识算法的辨识准确度较高,可以为伤员心肺复苏按压过程提供指导,辅助医学救援人员实施高效率的检伤和救治工作。

Objective To propose an new identification algorithm based on stacked autoencoder(SAE)deep neural network to enable accurate recognition of the return of spontaneous circulation during compression resuscitation and to guide the initiation and cessation of chest compressions.Methods Ten experimental pigs were subjected to ventricular fibrillation induction and defibrillation to collect data containing the return phase of the spontaneous circulation,during which changes in pulse wave signals were continuously recorded.After data collection the pulse waves were processed and an energy map was obtained,which was modelled and predicted based on an SAE deep neural network to achieve accurate identification of the spontaneous circulation.Results The experimental results showed that the SAE deep neural network-based algorithm had the average accuracy being 95.0%and 86.5%in identifying the return of spontaneous circulation in the absence and presence of chest compressions,respectively,during the cardiopulmonary resuscitation for cardiac arrest.Conclusion The algorithm proposed with high identification accuracy provides guidance during casualty cardiopulmonary resuscitation compression and assists medical rescuers for casualty triage and treatment.