Towards automated phenotyping in epilepsy

美国有500多万儿童和成年人被诊断为癫痫或癫痫 无序。然而，癫痫的治疗选择仍然不足，因为许多患者患有 不受控制的癫痫发作和治疗的负面副作用。速度更快的主要障碍 新的抗惊厥疗法的发展是严格的临床前癫痫研究通常 需要劳动密集型且昂贵的全天候视频-EEG对癫痫发作的监测依赖于主观 人类观察者对癫痫发作表型的评分(例如广泛使用的瑞辛量表 行为发作)。我们建议测试是否有可能执行客观、廉价和自动化的 获得性和遗传性癫痫小鼠不同模型的表型。这种方法依赖于 最近认识到，鼠标行为是在亚秒时间尺度上以刻板印象的模块构成的， 都是按照特定的规则安排的。这些特有的行为模块，以及 通过结合3D成像和机器学习(ML)，可以在没有观察者偏见的情况下识别它们 分析方法。我们建议将这种新的ML辅助3D视频分析技术应用于癫痫 研究，以测试它是否可以用于识别慢性颞叶癫痫(TLE)小鼠在 两种不同的实验性TLE模型的发作间期和发作期，以及不同的实验 条件。此外，我们还将测试该方法是否能够自动检测到 重度儿童癫痫遗传模型中的显性癫痫小鼠(纯合子电压门控钠 通道β-亚单位SCN1B-/-敲除小鼠)，但也区分看似正常，非癫痫， SCN1B/-野生型对照杂合子小鼠。我们预计这些结果将会有一个 通过展示自动化、客观、 独立于用户，廉价的获得性和遗传性癫痫表型分析。

Over 5 million children and adults in the United States have had a diagnosis of epilepsy or a seizure disorder. However, treatment options for the epilepsies remain inadequate, because many patients suffer from uncontrolled seizures and from the negative side effects of treatment. A major obstacle to the faster development of new anti-convulsant therapies is the fact that rigorous preclinical epilepsy research typically requires labor-intensive and expensive 24/7 video-EEG monitoring of seizures that rests on the subjective scoring of seizure phenotypes by human observers (as exemplified by the widely used Racine scale of behavioral seizures). We propose to test if it is possible to perform objective, inexpensive and automated phenotyping of mice in various mouse models of acquired and genetic epilepsies. The approach rests on the recent recognition that mouse behaviors are structured in stereotyped modules at sub-second timescales that are arranged according to specific rules. These characteristic behavioral modules, and the transitions between them, can be identified without observer bias by combined 3D imaging and machine learning (ML) -assisted analytic methods. We propose to adopt this novel ML-assisted 3D video analysis technology to epilepsy research, in order to test if it can be used to identify mice with chronic temporal lobe epilepsy (TLE) during inter-ictal and ictal periods in two distinct experimental TLE models, and under various experimental conditions. In addition, we will also test whether the approach is able to automatically detect not only the overtly epileptic mice in a genetic model of severe childhood epilepsy (homozygous voltage-gated sodium channel β-subunit SCN1B-/- knock-out mice), but also distinguish the seemingly normal, non-epileptic, SCN1B+/- heterozygous mice from the wild-type controls. We anticipate that these results will have a potentially transformative effect on the field by demonstrating the feasibility and power of automated, objective, user-independent, inexpensive analysis of acquired and genetic epilepsy phenotypes.