本文针对传统开环迷走神经刺激及其磁铁模式治疗药物难治性癫痫存在的局限性，提出了基于心电实现闭环迷走神经刺激，增加对特定类型癫痫发作的电刺激，以提高迷走神经刺激的治疗效果。 本文首先采用了一种QRS波群检测算法，得到心电信号的RR间隔和瞬时心率。通过在MIT公开数据库中选取7位患者的部分性癫痫发作后心率振荡的数据进行分析，结果表明与标准RR间隔的平均误差低至0.78毫秒。同时还提出了一种基于心率的癫痫发作检测算法，并在相同数据集中发现，伴随心动过速的9次发作均被成功检测到，敏感度达100%。结果表明，QRS检测算法可以准确地检测出心率，癫痫发作检测算法具有很高的灵敏度和检测速率。随后，通过临床试验，对比不同设备同步采集下的心电信号的QRS波特征、数量，并计算RR间隔和瞬时心率，验证了闭环迷走神经刺激设备在体内心电感知功能的准确性。在自动刺激临床试验中， 5例癫痫发作时心率明显升高的药物难治性癫痫患者筛选入组，通过收集不同状态下的心电信号，对比运动和癫痫发作对于心率的影响差异，在优先保证敏感性指标的前提下，完成了患者个体最适自动刺激触发阈值的设定。植入闭环迷走神经刺激设备后，在癫痫监测单元（EMU）内仅开启自动刺激模式，记录长程视频脑电心电数据，通过对比癫痫发作的实际时间和自动刺激模式触发记录时间，发现发作检测的敏感度可达100%，但假阳性也达到1.28 次/小时。临床结果表明，基于心电的闭环迷走神经刺激具有极高的敏感性和较高的假阳性。最后，针对假阳性刺激原因，提出了结合心率变异性指标和心率平均变化率的二次判断优化算法，选用对真实发作、假发作状态具有显著区分的特征作为二次判断的依据，并设置个体最适特征和阈值，在保证敏感性不降低的前提下实现了假阳性平均降低36%，进一步提高闭环迷走神经刺激的精准干预。综上，本文工作是国内首次基于心电的闭环迷走神经刺激疗法治疗药物难治性癫痫的分析和应用，从心率检测、癫痫发作检测、自动刺激输出以及优化假阳性刺激算法四个方面实现了闭环迷走神经刺激的系统性软件设计，具有提高癫痫发作的自动化精准干预的临床价值。
Focusing on tackling the limitations of traditional open-loop vagus nerve stimulation (VNS) and its magnet mode for drug-resistant epilepsy (DRE) treatment, this article proposes a closed-loop VNS therapy based on electrocardiogram (ECG). The proposed closed-loop VNS therapy automatically triggers stimulation to specific types of epileptic seizures in order to improve the therapeutic effect of VNS.Firstly, a QRS complex detection algorithm was introduced to obtain the RR interval and instantaneous heart rate of the ECG signal. Analysis of heart rate oscillation data after partial epileptic seizures in seven patients from the database of MIT Laboratory for Computational Physiology was carried out, and the average error of standard RR interval was as low as 0.78 milliseconds Additionally, a heart rate-based seizure detection algorithm was developed, and it was found in the same dataset that all nine seizures accompanied by tachycardia were successfully detected, with a sensitivity of 100%. These results demonstrate the accuracy of the QRS detection algorithm in detecting heart rate and the high sensitivity and detection rate of the seizure detection algorithm.Subsequently, a clinical trial was conducted to compare the features and quantity of QRS complex in ECG signals collected by different devices at the same time. The RR interval and instantaneous heart rate were calculated to verify the accuracy of the closed-loop VNS device‘s ECG sensing function in vivo. Five DRE patients with significantly increased heart rate during seizures were selected for automatic stimulation clinical trials. By collecting ECG signals in different states and comparing the difference in heart rate between movement and seizures, individualized optimal automatic stimulation trigger thresholds were set, prioritizing sensitivity. After implantation of the closed-loop VNS device, only automatic stimulation mode was turned on during the epilepsy monitoring unit(EMU), and long-term video electroencephalogram and ECG data were recorded. By comparing the actual time of seizures and the recorded time of automatic stimulation triggered, it was found that the sensitivity of seizure detection reached 100%, but the false positive rate was 1.28 times/hour. Clinical results show that the ECG-based closed-loop VNS has high sensitivity and relatively high false positive rate.Finally, an optimization algorithm combining heart rate variability indicators and average heart rate change was proposed to address the false positive stimulation issue. We first identified key features that helped us distinguish between true and false seizure states. Individual optimal features and thresholds were then set to achieve an average reduction in false positive rate of 36% without reducing sensitivity, further improving the accuracy of closed-loop VNS intervention.In summary, this study is the first in China to analyze and apply ECG-based closed-loop VNS therapy for DRE, and has realized a systematic software design of closed-loop VNS in four aspects: heart rate detection, seizure detection, automatic stimulation output, and false positive stimulation algorithm optimization. It has significant clinical value in improving the automated precision intervention of epilepsy.
