Active Probing and Sleep State Modeling for Seizure Prediction

用于癫痫发作预测的主动探测和睡眠状态建模

• 批准号: 8113291
• 负责人: Madineh Sedigh-Sarvestani
• 金额: $ 3.48万
• 依托单位: PENNSYLVANIA STATE UNIVERSITY, THE
• 依托单位国家: 美国
• 财政年份: 2010
• 资助国家: 美国
• 起止时间: 2010-07-07 至 2013-07-06
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/8113291
• 关键词: Accounting; Algorithms; Animals; Assimilations; Behavior; Behavioral; Brain; Clinical; Cognitive; Computer Simulation; Data; Dependence; Development; Devices; Electroencephalography; Emotional; Epilepsy; Failure; Frequencies; Goals; Implant; Intervention; Intractable Epilepsy; Investigation; Laboratory Finding; Link; Measurement; Methods; Modeling; Monitor; Output; Patients; Performance; Persons; Publishing; Quality of life; REM Sleep; Rattus; Rodent; Seizures; Sensitivity and Specificity; Sleep; Staging; Symptoms; System; Temporal Lobe Epilepsy; Tetanus Toxin; The Sun; Work; base; clinically relevant; design; electric field; improved; insight; non rapid eye movement; novel; prevent; relating to nervous system; research study; response; success; theories; vigilance

Much effort is currently directed towards the design of a responsive (closed-loop) stimulation device for the treatment of epilepsy (Osorio 2001, Sun 2008).The ideal intervention system would be able to prevent the occurrence of a seizure before the onset of behavioral and clinical symptoms, so as to impose minimum cognitive and emotional effects on the patient. However, current published seizure prediction algorithms are still too limited to be incorporated into these closed-loop devices. The aim of this project is to develop a sensitive and specific seizure prediction algorithm. We propose to use novel measurements of brain dynamics to improve the performance of current algorithms. Clinical and laboratory findings support the existence of a 'preseizure' state. In theory, such a state should be detectable. Despite initial promising results, detecting the preseizure state has proven to be a challenging task, in part due to the confounding effect of state of vigilance (Wake, non-rapid eye movement sleep, rapid eye movement sleep) on seizure dynamics (Schelter 2006). In reviewing the state-of-the-art in seizure prediction in 2007, Mormann et al. (Mormann 2007) used careful statistical analysis of published results to document that despite some evidence of success, seizure prediction was still quite difficult and they attributed a failure to account for SOV as one of several problems. Despite these warnings, little effort has been put forth to incorporate SOV in published seizure prediction algorithms. Furthermore, most seizure-prediction algorithms utilize passive measurements of brain dynamics, such as spontaneous EEG. We propose here to incorporate SOV as well as active measurements of brain dynamics as additional feature types to be used as input to the seizure prediction algorithm: We will probe the brain with polarizing low-frequency electric field (PLEF) stimulation and record the response during the preseizure (period immediately before seizure onset) and interictal (period well before seizure onset, between two seizures) periods. We will extract several characterizing feature sets from these responses and use them to distinguish the preseizure state. We will conduct these experiments in the rat tetanus toxin model of temporal lobe epilepsy. In SA1, we will develop a seizure predictor based on several different features extracted from the spontaneous EEG, including the state of vigilance. We will use this passive predictor as our working baseline. In SA2, we will implement an active probing paradigm to monitor the dynamic state of the brain by recording neural responses to PLEF between and before seizures. We will use these responses, along with SOV, as input features to augment our predictive algorithm. In SA3, we will use computational modeling to gain insight to the underlying dynamics of sleep and the intimate link between sleep state and seizure (Dinner 2002). Wewill use algorithms developed under this aim to reconstruct 'hidden' variables in a dynamical model of sleep. These variables will be evaluated as potential input features for our seizure prediction algorithm.

目前，许多努力都用于设计一种响应式（闭环）刺激装置，以治疗癫痫病（Osorio 2001，Sun 2008）。理想的干预系统将能够防止在行为和临床症状发作之前发生癫痫发作，从而对患者施加最小的认知和情绪影响。但是，当前已发布的癫痫发作预测算法仍然太有限，无法将其纳入这些闭环设备。该项目的目的是开发一种敏感且特定的癫痫发作预测算法。我们建议使用大脑动力学的新型测量结果来改善当前算法的性能。临床和实验室发现支持“ preseizure”状态的存在。从理论上讲，应检测到这种状态。尽管最初有希望的结果，但检测到率状态已被证明是一项具有挑战性的任务，部分原因是警惕状态的混淆（唤醒，非比式眼动睡眠，快速眼动睡眠）对癫痫发作动力学（Schelter 2006）。 Mormann等人在审查2007年癫痫发作预测的最新预测时。 （Mormann 2007）对已发表的结果进行了仔细的统计分析，以证明尽管有一些成功的证据，但癫痫发作预测仍然很困难，他们归因于未能将SOV视为几个问题之一。尽管有这些警告，但几乎没有努力将SoV纳入已发表的癫痫发作预测算法中。此外，大多数癫痫发作预测算法都利用了对脑动力学的无源测量，例如自发的脑电图。我们在这里提议将SOV和主动测量大脑动力学作为其他特征类型，以作为癫痫发作预测算法的输入：我们将用极化的低频电场（PLEF）刺激探测大脑，并记录preseizure（在癫痫发作之前）和疗程（持续期间）（隔离之前）（在隔离之前）（两次seire seire seire seire seire seire seire seire），并记录响应。我们将从这些响应中提取几个特征集，并使用它们来区分perseizure态。我们将在颞叶癫痫的大鼠破伤风毒素模型中进行这些实验。在SA1中，我们将根据从自发的脑电图中提取的几种不同特征（包括警惕状态）开发癫痫发作预测变量。我们将使用此被动预测指标作为我们的工作基线。在SA2中，我们将通过记录癫痫发作和之前对PLEF的神经反应来监测大脑的动态状态，以监测大脑的动态状态。我们将使用这些响应以及SOV作为输入功能来增强我们的预测算法。在SA3中，我们将使用计算建模来洞悉睡眠的潜在动态以及睡眠状态与癫痫发作之间的紧密联系（晚餐2002）。我们将使用此目的开发的算法，以在动态的睡眠模型中重建“隐藏”变量。这些变量将评估为我们的癫痫发作预测算法的潜在输入特征。