PDF(Pubmed)
Abstract：
Epilepsy is characterized by recurrent, unprovoked seizures. Accurate prediction of seizure occurrence has long been a clinical goal since this would allow to optimize patient treatment, prevent injuries due to seizures, and alleviate the patient burden of unpredictability. Advances in implantable electroencephalographic (EEG) devices, allowing for long-term interictal EEG recordings, have facilitated major progress in this field. Recently, it has been discovered that interictal brain activity demonstrates circadian and multi-dien cycles that are strongly aligned, or phase locked, with seizure risk. Thus, cyclical brain activity patterns have been used to forecast seizures. However, in the effort to develop a clinically useful EEG based seizure forecasting system, challenges remain. Firstly, multiple EEG features demonstrate cyclical patterns, but it remains unclear which feature is best suited for predicting seizures. Secondly, the technology for long-term EEG recording is currently limited in both spatial and temporal sampling resolution. In this study, we compare five established EEG metrics:synchrony, spatial correlation, temporal correlation, signal variance which have been motivated from critical dynamics theory, and interictal epileptiform discharge (IED) which are a traditional marker of seizure propensity. We assess their effectiveness in detecting 24-h and seizure cycles as well as their robustness under spatial and temporal subsampling. Analyzing intracranial EEG data from 23 patients, we report that all examined features exhibit 24-h cycles. Spatial correlation, signal variance, and synchrony showed the highest phase locking with seizures, while IED rates were the lowest. Notably, spatial and temporal correlation were also found to be highly correlated to each other, as were signal variance and IED-suggesting some features may reflect similar aspects of cortical dynamics, whereas others provide complementary information. All features proved robust under subsampling, indicating that the dynamic properties of interictal activity evolve slowly and are not confined to specific brain regions. Our results may aid future translational research by assisting in design and testing of EEG based seizure forecasting systems.
摘要：
癫痫的特点是反复发作，无缘无故的癫痫发作。准确预测癫痫发作的发生长期以来一直是一个临床目标，因为这将允许优化患者治疗。防止癫痫发作造成的伤害，减轻患者的不可预测性负担。植入式脑电图（EEG）设备的进展，允许长期的发作间脑电图记录，促进了这一领域的重大进展。最近,已经发现，发作间的大脑活动表现出强烈一致的昼夜节律和多周期，或锁相，有癫痫发作的风险。因此，周期性的大脑活动模式已被用来预测癫痫发作。然而,为了开发一种临床上有用的基于脑电图的癫痫发作预测系统，挑战依然存在。首先,多个脑电图特征表现出周期性模式，但尚不清楚哪种特征最适合预测癫痫发作。其次,用于长期脑电图记录的技术目前在空间和时间采样分辨率方面都受到限制。在这项研究中,我们比较了五种既定的脑电图指标：同步性，空间相关性，时间相关性，来自临界动力学理论的信号方差，和癫痫间期癫痫样放电（IED）是癫痫发作倾向的传统标志。我们评估了它们在检测24小时和癫痫发作周期中的有效性，以及在空间和时间子采样下的鲁棒性。分析23例患者的颅内脑电图数据，我们报告所有检查的特征都表现出24小时周期。空间相关性，信号方差，同步性表现出与癫痫发作的最高相位锁定，而IED率最低。值得注意的是,时空相关性也被发现是高度相关的，信号方差和IED-提示一些特征可能反映了皮质动力学的相似方面，而其他人提供补充信息。在子采样下，所有特征都被证明是稳健的，表明发作间活动的动态特性进化缓慢，并不局限于特定的大脑区域。我们的结果可以通过协助设计和测试基于EEG的癫痫发作预测系统来帮助未来的转化研究。
