Algorithm for the Real-Time Detection of Absence Seizures from Oculometric Data

根据眼科数据实时检测失神发作的算法

• 批准号: 10267036
• 负责人: Rachel Kuperman
• 金额: $ 13.69万
• 依托单位: EYSZ, INC.
• 依托单位国家: 美国
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-09-30 至 2022-11-30
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10267036
• 关键词: Absence Epilepsy; Activities of Daily Living; Address; Adult; Advanced Development; Age; Algorithmic Software; Algorithms; Anticonvulsants; Award; Awareness; Blinking; Capital; Cessation of life; Childhood; Chronic; Clinic; Clinical Research; Clinical Trials; Complex; Data; Data Set; Databases; Decision Making; Detection; Development; Devices; Diagnosis; Early Diagnosis; Early treatment; Electrodes; Electroencephalogram; Epilepsy; Eye Movements; Focal Seizure; Foundations; Frequencies; Future; General Population; Glass; Gold; Impairment; Individual; Letters; Location; Machine Learning; Medical; Monitor; Morbidity - disease rate; Morphologic artifacts; Motion; Movement; Outpatients; Patients; Performance; Phase; Physicians; Pupil; Quality of life; Resolution; Risk; Seizures; Shark; Small Business Innovation Research Grant; Specificity; Statistical Data Interpretation; Statistical Methods; Symptoms; Technology; Testing; Therapeutic; Time; Tonic - clonic seizures; Training; Treatment Protocols; Work; algorithm development; base; commercialization; detection platform; digital health; disability; experience; high risk; improved; improved outcome; innovation; large datasets; machine learning method; mortality; mortality risk; nervous system disorder; premature; prospective; software systems; statistical and machine learning; visual tracking; wearable device

Abstract Eysz, Inc. is developing an algorithm and software solutions to reliably and affordably detect seizures in an ambulatory setting using existing smart glass technologies. In a proof-of-concept study, Eysz was able to detect >75% of all absence seizures longer than 10 s in duration using only oculometric variables (e.g., pupil size, pupil location, eccentricity, blink frequency) detected using off-the-shelf eye-tracking technology. Eysz seeks to build on this finding by developing and commercializing highly sensitive and specific seizure detection algorithms using eye-movement data as input, with eventual expansion to additional seizure types. This technology has the potential to transform the detection and treatment of seizures for those with epilepsy, one of the most common neurological disorders worldwide. Timely treatment can reduce the chance of additional seizures by half, making early detection and treatment critical. Unfortunately, detection and diagnosis can be difficult using current technologies, especially in types of epilepsy with few observable symptoms such as absence seizures. The gold standard for detecting and characterizing seizure activity is electroencephalogram (EEG) monitoring with video and subsequent review by a trained clinician, but this does not translate well to the outpatient setting. While attempts to develop ambulatory EEGs have been made, these have significant drawbacks, including poor patient acceptability, poor detection capability, and continued reliance on asynchronous review. Additional non-EEG- based motion detection devices are limited to tonic-clonic seizures, which are responsible for a small fraction of all seizure activity. Thus, there is a critical need to reliably detect seizures outside of the clinic to provide physicians with necessary information to guide therapeutic decision making. To address this need, Eysz is developing a digital health platform that leverages existing eye tracking technology to meet this significant unmet gap in the market and is technically feasible, capital-efficient, robust, and innovative. Eysz plans to use existing smart glass technology to export the necessary oculometric data to be analyzed by our seizure detection algorithm. We will also build out databases, software systems, and user interfaces enabling the resulting data to be stored in the cloud and visualized/analyzed by physicians. In this Phase I SBIR, Eysz will advance the development of the seizure detection algorithms by: 1) obtaining oculometric video and EEG data on ≥100 absence seizures from multiple patients, and 2) using ML and statistical methods to optimize an algorithm for identifying absence seizures using eye-tracking data, with a target sensitivity of 85% and specificity of 90%. Lessons learned from this study will be applied (with different training sets) to additional seizures types, such as focal impaired awareness (formerly called complex partial) seizures, the most prevalent seizure type in adults. This work is of critical importance to the field, as demonstrated by support from the Epilepsy Foundation and receipt of both the judges' and people's choice awards in the Epilepsy Foundation's 8th Annual Shark Tank Competition.

摘要 Eysz公司正在开发一种算法和软件解决方案，以可靠和经济地检测癫痫发作， 使用现有的智能玻璃技术。在一项概念验证研究中，Eysz能够检测到 >75%的失神发作持续时间超过10 s，仅使用视力测量变量（例如，瞳孔大小 位置、偏心率、眨眼频率）。Eysz试图建立 通过开发和商业化高度敏感和特异性的癫痫发作检测算法， 使用眼球运动数据作为输入，最终扩展到其他癫痫发作类型。这项技术有 有可能改变癫痫患者癫痫发作的检测和治疗，这是最常见的 世界范围内的神经系统疾病。及时治疗可以减少额外发作的机会一半，使 早期发现和治疗至关重要。不幸的是，检测和诊断可能是困难的，使用电流 这类药物可以用于治疗癫痫病，特别是那些几乎没有明显症状的癫痫病，如失神发作。黄金 用于检测和表征癫痫发作活动的标准是具有视频的脑电图（EEG）监测 随后由经过培训的临床医生进行复查，但这并不能很好地转化为门诊设置。而 已经进行了开发动态EEG的尝试，但是这些具有显著的缺点，包括患者不好， 可接受性，检测能力差，继续依赖异步审查。其他非EEG- 基于运动检测的设备仅限于强直阵挛发作，这是造成小部分癫痫发作的原因。 所有的癫痫发作因此，迫切需要在诊所外可靠地检测癫痫发作，以提供诊断。 医生提供必要的信息，以指导治疗决策。为了满足这一需求，Eysz 开发数字健康平台，利用现有的眼动追踪技术来满足这一重大未满足的需求 这是一个技术上可行、资本效率高、稳健和创新的市场。Eysz计划利用现有的 智能玻璃技术输出必要的视力测量数据，以供我们的检获检测进行分析 算法我们还将建立数据库、软件系统和用户界面，使生成的数据能够 存储在云中并由医生可视化/分析。在第一阶段SBIR中，Eysz将推进 通过以下方式开发癫痫发作检测算法：1）获得≥100例患者的视力视频和EEG数据 来自多个患者的失神发作，以及2）使用ML和统计方法来优化用于 使用眼动追踪数据识别失神发作，目标灵敏度为85%，特异性为90%。 从这项研究中吸取的经验教训将（通过不同的培训集）应用于其他癫痫发作类型，例如 局灶性意识受损（以前称为复杂部分性）癫痫发作，是成人中最常见的癫痫发作类型。 这项工作对该领域至关重要，癫痫基金会和 在癫痫基金会的第八届年度鲨鱼坦克中获得评委和人民的选择奖 竞争