Refining neurophysiological biomarkers of epilepsy using deep learning to guide pediatric epilepsy surgery

利用深度学习完善癫痫的神经生理学生物标志物来指导小儿癫痫手术

• 批准号: 10664790
• 负责人: Hiroki Nariai
• 金额: $ 24.21万
• 依托单位: UNIVERSITY OF CALIFORNIA LOS ANGELES
• 依托单位国家: 美国
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-04-01 至 2028-03-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10664790
• 关键词: Affect; Algorithms; Area; Behavioral; Biological; Biological Markers; Brain; Brain region; Characteristics; Child; Childhood; Chronic; Clinical; Clinical Research; Clinical Trials; Data; Electroencephalography; Epilepsy; Evaluation; Excision; Freedom; Frequencies; Goals; Hand; High Frequency Oscillation; Image; Joints; K-Series Research Career Programs; Knowledge; Label; Language; Link; Maps; Mentors; Mentorship; Methods; Microelectrodes; Modeling; Morbidity - disease rate; Morphology; Neocortex; Neurons; Operative Surgical Procedures; Outcome; Partial Epilepsies; Pathologic; Pharmaceutical Preparations; Physicians; Physiological; Postoperative Period; Property; Prospective cohort; Research; Research Personnel; Resistance; Retrospective cohort; Scientist; Seizures; Signal Transduction; Site; Synaptic Potentials; Testing; Time; Training; Vision; Work; biomarker identification; brain abnormalities; brain tissue; childhood epilepsy; cohort; cortex mapping; deep learning; deep learning algorithm; deep learning model; dexterity; feasibility testing; large datasets; mortality; neocortical; neurophysiology; neuroregulation; novel; postsynaptic; predictive modeling; preservation; secondary analysis; signal processing; skill acquisition; skills; statistics; theories; tool; two-dimensional

PROJECT SUMMARY/ABSTRACT Dr. Hiroki Nariai is a pediatric epileptologist/clinical neurophysiologist whose long-term goal is to be a leading physician-scientist in pediatric epilepsy, using key biomarkers to effectively treat children with epilepsy and reduce their mortality and morbidity. In this project, Dr. Nariai proposes to study medication-resistant focal epilepsy in children by integrating computational electroencephalogram (EEG) analysis, deep learning, and advanced statistics to investigate and validate high-frequency oscillations (HFOs)—a promising spatial biomarker of the epileptic brain. More than one-third of children with epilepsy are resistant to medications and are therefore potential candidates for epilepsy surgery. To achieve postoperative seizure freedom, one must remove or disrupt the epileptogenic zone (EZ), defined as the brain area that is indispensable for generating seizures, while preserving the eloquent cortex (EC), defined as the brain area that controls essential functions. Thus, identifying biomarkers that accurately localize and discriminate EZ from EC will be groundbreaking. HFOs are recorded via intracranial EEG as short bursts of high-frequency neuronal activity and are often observed in EZ. However, the major challenge is that physiological HFOs generated by healthy brain tissue complicate the clinical interpretation of HFOs. Therefore, there is a critical need to distinguish between pathological and physiological HFOs. Dr. Nariai hypothesizes that deep learning-based algorithms can distinguish pathological and physiological HFOs based on subtle morphological features linked to specific biological mechanisms. Through this K23 career development award, Dr. Nariai proposes to accomplish the following training goals: (1) acquire skills in an advanced computational EEG analysis to enable customized quantification of HFOs in a large dataset, (2) gain knowledge of the theory of deep learning and skills in its application in EEG signal processing to enable morphological assessment of HFOs, and (3) develop proficiency in advanced statistics in clinical research to validate prediction models and gain knowledge in clinical trials. Under the joint mentorship of leading researchers led by Dr. Jerome Engel, Jr., at UCLA, Dr. Nariai will build deep learning-based models in a large retrospective cohort to define HFOs expressed in EZ (eHFOs) to represent pathological HFOs. In addition, HFOs expressed in EC (ecHFOs) will be defined to represent physiological HFOs. The trained classifier will be analyzed to obtain the computational definition of eHFOs and ecHFOs. Along with demonstrating that real-time HFO analysis is feasible in a prospective cohort, eHFOs and ecHFOs will be analyzed to prove that HFOs can localize and discriminate EZ from EC. Dr. Nariai has shown preliminary results supporting the feasibility of his proposed approach. Completing the proposed goals will provide significant progress toward utilizing HFOs as a clinically useful spatial biomarker of the epileptic brain to guide epilepsy surgery, which Dr. Nariai plans to pursue as an R01 project.

项目概要/摘要 Hiroki Nariai 博士是一名儿科癫痫病学家/临床神经生理学家，其长期目标是成为领先的 小儿癫痫领域的医师科学家，使用关键生物标志物有效治疗儿童癫痫和 降低他们的死亡率和发病率。在这个项目中，Nariai 博士建议研究耐药性局灶性 通过整合计算脑电图 (EEG) 分析、深度学习和 用于研究和验证高频振荡 (HFO) 的高级统计数据——一种有前途的空间 癫痫脑的生物标志物。超过三分之一的癫痫儿童对药物有抵抗力 因此是癫痫手术的潜在候选者。为了实现术后无癫痫发作，必须 去除或破坏致痫区（EZ），致痫区定义为产生癫痫所不可缺少的大脑区域。 癫痫发作，同时保留口才皮层（EC），其定义为控制基本功能的大脑区域。 因此，识别能够准确定位和区分 EZ 和 EC 的生物标志物将具有开创性。氢氟酸 通过颅内脑电图记录为高频神经元活动的短爆发，并且经常在以下情况中观察到： EZ。然而，主要的挑战是健康脑组织产生的生理性 HFO 使 HFO 的临床解释。因此，区分病理性和病理性是非常必要的。 生理性 HFO。 Nariai 博士假设基于深度学习的算法可以区分病理性的 以及基于与特定生物机制相关的微妙形态特征的生理性 HFO。 通过这个K23职业发展奖，Nariai博士建议实现以下培训目标：（1） 获得高级计算脑电图分析技能，以实现大规模 HFO 的定制量化 数据集，（2）获得深度学习理论知识及其在脑电信号处理中的应用技能 实现 HFO 的形态学评估，以及 (3) 提高临床高级统计的熟练程度 研究验证预测模型并获得临床试验知识。在领导的共同指导下 由加州大学洛杉矶分校 Jerome Engel, Jr. 博士领导的研究人员和 Nariai 博士将在大规模中构建基于深度学习的模型 回顾性队列定义 EZ 中表达的 HFO（eHFO）来代表病理性 HFO。此外，重油 以 EC (ecHFO) 表示的物质将被定义为代表生理性 HFO。将分析训练好的分类器 获得 eHFO 和 ecHFO 的计算定义。除了展示实时 HFO 分析在前瞻性队列中是可行的，将对 eHFO 和 ecHFO 进行分析，以证明 HFO 可以定位 并区分 EZ 和 EC。 Nariai 博士已经展示了支持其提议可行性的初步结果 方法。完成拟议的目标将为利用 HFO 作为临床药物取得重大进展 癫痫大脑的有用空间生物标志物可指导癫痫手术，Nariai 博士计划将其作为 R01项目。