Predicting Epilepsy Surgery Outcomes Using Neural Network Architecture

使用神经网络架构预测癫痫手术结果

• 批准号: 10649724
• 负责人: Leonardo F Bonilha
• 金额: $ 81.28万
• 依托单位: EMORY UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2019
• 资助国家: 美国
• 起止时间: 2019-08-15 至 2024-05-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10649724
• 关键词: Ablation; Anterior Temporal Lobectomy; Anticonvulsants; Area; Assessment tool; Basic Science; Brain; Brain Mapping; Caring; Clinical; Clinical Data; Common Data Element; Contralateral; Counseling; Data; Data Collection; Detection; Diffusion; Diffusion Magnetic Resonance Imaging; Electrocoagulation; Electroencephalography; Ensure; Epilepsy; Evaluation; Excision; Failure; Fiber; Freedom; Functional Magnetic Resonance Imaging; Functional disorder; Future; Hippocampus; Image; Individual; Intractable Epilepsy; Knowledge; Location; Machine Learning; Magnetic Resonance Imaging; Maps; Measures; Medial; Methods; Modality; Modeling; Modernization; Neurobiology; Neurons; Operative Surgical Procedures; Outcome; Pathologic; Pathway interactions; Patients; Pharmaceutical Preparations; Phenotype; Predictive Value; Procedures; Prospective Studies; Prospective cohort; Publishing; Refractory; Reproducibility; Research; Resistance; Rest; Risk; Scalp structure; Seizures; Severities; Specificity; Structure; Temporal Lobe Epilepsy; Testing; Thalamic structure; Tonic-Clonic Epilepsy; United States National Institutes of Health; Validation; base; computerized tools; connectome; experience; gray matter; hippocampal atrophy; imaging modality; improved; millimeter; multimodality; neural network; neural network architecture; neuroimaging; neuronal circuitry; novel strategies; outcome prediction; pharmacologic; predictive modeling; prospective; recruit; response; standard of care; success; surgery outcome; tractography; white matter

Abstract Temporal lobe epilepsy (TLE) is one of the most common forms of pharmacologically resistant epilepsy. The resection or ablation of medial temporal structures can be curative for many patients. Unfortunately, approximately one third of patients who undergo TLE surgery continue to have disabling seizures post- procedurally. The reasons for suboptimal outcomes are not well understood and therefore constitute a very important knowledge gap in epilepsy care. A better understanding of this difference in surgical response phenotype could be used to improve surgical planning, treatment, outcome prediction and counseling. Promising preliminary studies suggest that TLE surgical outcomes can be inferred by neuroimaging computational tools assessing the cumulative degree of abnormalities in the topological organization of structural networks involving limbic and extra-limbic regions. Nonetheless, network abnormalities are not routinely or systematically used and quantified in the pre-surgical evaluation of epilepsies, and their assessment requires refinement and further validation. The purpose of this proposal is to perform a prospective study to test the hypothesis that the degree of limbic and extra-limbic network abnormalities in TLE, systematically assessed using a connectome approach based on optimized diffusion MRI (dMRI), can be used to predict and better understand epilepsy surgery outcomes. This hypothesis builds on the well-defined basic science and neurobiological premises that epilepsy is associated with pathological alterations in networks that are related to seizure onset and seizure propagation. Importantly, network abnormalities are not visible on routine MRI, but their detection using connectomes constitutes a modern approach to quantifying the location and magnitude of “lesional epilepsy,” where broad computational network abnormalities imply worse outcomes. We will prospectively gather clinical and imaging data at six epilepsy centers using the NIH epilepsy common data elements. This project will be fundamentally based on standard of care data, thus minimizing the burden of extra data collection and ensuring feasibility. Furthermore, this project will be embedded in the ENIGMA-Epilepsy framework, which is a collaborative platform for clinical and neuromaging multi-center research. Specific Aim 1 will define the accuracy, reproducibility, and predictive values of the pre-surgical dMRI tractography connectome model towards surgical results in TLE. We will perform hypothesis-driven tests of specific limbic and extra-limbic networks in relationship with clinical data and surgical outcomes. Specific Aim 2 will test if the neuroimaging-clinical outcome model can be further improved with advanced diffusion methods (multi-shell diffusional kurtosis imaging), resting state functional MRI networks, or a multimodal approach. We believe that this research will have an important impact on our understanding of the mechanisms related to TLE treatment.

摘要 颞叶癫痫（Temporal lobe epilepsy，TLE）是一种最常见的难治性癫痫。的 内侧颞部结构的切除或消融对于许多患者是治愈性的。不幸的是， 大约三分之一的接受TLE手术的患者在术后仍有致残性癫痫发作， 程序上。次优结果的原因尚不清楚，因此构成了一个非常 癫痫治疗中的重要知识差距。更好地理解手术反应的这种差异 表型可用于改善手术计划、治疗、结果预测和咨询。 有希望的初步研究表明，颞叶癫痫手术的结果可以推断神经影像学 计算工具评估的累积程度异常的拓扑组织 涉及边缘和边缘外区域的结构网络。尽管如此，网络异常并不是 常规或系统地用于癫痫术前评估并进行量化， 评估需要完善和进一步验证。本提案的目的是执行一项 一项前瞻性研究，以检验以下假设： 使用基于优化扩散MRI（dMRI）的连接体方法系统评估TLE，可以 用于预测和更好地了解癫痫手术结果。这一假设建立在定义明确的 基础科学和神经生物学的前提，癫痫与病理改变有关， 与癫痫发作和癫痫传播相关的网络。重要的是，网络异常不是 在常规MRI上可见，但使用连接体的检测构成了一种现代方法来量化 “病灶性癫痫”的位置和程度，其中广泛的计算网络异常意味着更严重的 结果。我们将前瞻性地收集六个癫痫中心的临床和影像学数据， 常见的数据元素。该项目将从根本上基于护理标准数据，从而最大限度地减少 额外的数据收集负担和确保可行性。此外，该项目将被嵌入到 ENIGMA-癫痫框架，这是一个临床和神经成像多中心协作平台 research.具体目标1将定义术前dMRI的准确性、再现性和预测值 纤维束成像连接体模型对TLE手术结果的影响我们将进行假设驱动的测试， 与临床数据和手术结果相关的特定边缘系统和边缘系统外网络。具体目标2 将测试神经影像学-临床结果模型是否可以通过先进的扩散方法进一步改善 （多壳扩散峰度成像）、静息状态功能MRI网络或多模态方法。我们 我相信，这项研究将对我们理解有关的机制产生重要影响， TLE治疗。

项目成果

Seizure forecasting using machine learning models trained by seizure diaries.

• DOI: 10.1088/1361-6579/aca6ca
• 发表时间: 2022-12-14
• 期刊: Physiological measurement
• 影响因子: 3.2

High b-value diffusion tractography: Abnormal axonal network organization associated with medication-refractory epilepsy.

• DOI: 10.1016/j.neuroimage.2021.118866
• 发表时间: 2022-03
• 期刊: NeuroImage
• 影响因子: 5.7
• 作者: Gleichgerrcht E;Keller SS;Bryant L;Moss H;Kellermann TS;Biswas S;Marson AG;Wilmskoetter J;Jensen JH;Bonilha L
• 通讯作者: Bonilha L