Post-surgical resection mapping in epilepsy using convolutional neural networks

使用卷积神经网络绘制癫痫术后切除图

• 批准号: 10041126
• 负责人: Heath Pardoe
• 金额: $ 46.28万
• 依托单位: NEW YORK UNIVERSITY SCHOOL OF MEDICINE
• 依托单位国家: 美国
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-06-01 至 2022-05-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10041126
• 关键词: 3-Dimensional; Adverse effects; Anterior; Benchmarking; Brain; Brain region; Categories; Cognitive; Complex; Computer Analysis; Computer Hardware; Contralateral; Data; Data Set; Databases; Development; Epilepsy; Equilibrium; Excision; Goals; Health; Hippocampus (Brain); Histopathology; Hour; Image; Image Analysis; Individual; Intervention; Intractable Epilepsy; Investigation; Label; Literature; Location; MRI Scans; Machine Learning; Magnetic Resonance Imaging; Manuals; Maps; Measures; Medical; Medical center; Methods; National Institute of Neurological Disorders and Stroke; Network-based; New York; Normal tissue morphology; Operative Surgical Procedures; Outcome; Partial Epilepsies; Pathology; Patients; Pattern; Performance; Pharmaceutical Preparations; Physicians; Process; Radiology Specialty; Refractory; Research; Research Personnel; Resected; Resolution; Seizures; Structure; Surface; Techniques; Temporal Lobe; Tissues; Training; United States; Universities; base; brain abnormalities; brain surgery; brain tissue; cluster computing; cohort; computer framework; computerized tools; computing resources; convolutional neural network; deep learning; design; effective intervention; hippocampal sclerosis; improved; individual patient; learning strategy; neuroimaging; outcome prediction; parallel computer; standard of care; surgery outcome; treatment planning; treatment program

Project Summary Approximately one third of all individuals with epilepsy continue to have seizures despite treatment with anti- seizure medications. For these people, surgical removal of brain tissue can be a highly effective intervention to reduce or stop seizures. However, there is considerably variability in post-surgical seizure outcomes among individual patients, and the ability of physicians to predict who will benefit from surgery is limited. The location and extent of removed tissue, as well as neuroanatomical structures that are not surgically removed, are important factors that contribute to post-surgical outcomes. The goal of this proposal is to use convolutional neural networks, also known as deep learning, to map both the location and extent of surgically removed tissue on postsurgical MRI scans. The technique will also be used to automatically label brain regions that are spared during the surgical procedure. These computational tools will allow researchers to develop improved methods to predict postsurgical health outcomes. We will develop the automated method by training convolutional neural networks to identify brain regions on MRI scans obtained after epilepsy surgery at the New York University Langone Medical Center. CNNs have been specifically designed for the identification of complex spatial patterns in images and are likely to be well-suited to the identifications of changes in the brain following surgery. Recent developments in computer hardware and analysis methods mean that CNNs can now be applied to high resolution three-dimensional MRI scans. This project will leverage these recent developments in computational image analysis to improve our ability to predict outcomes following epilepsy surgery and therefore contribute to improved treatment for epilepsy patients.

项目摘要 大约三分之一的癫痫患者尽管接受了抗癫痫药物治疗， 癫痫药物对于这些人来说，手术切除脑组织是一种非常有效的干预措施， 减少或停止癫痫发作。然而，手术后癫痫发作的结果在以下人群中存在相当大的差异： 个体患者，医生预测谁将从手术中受益的能力有限。位置 和切除组织的范围，以及未手术切除的神经解剖结构， 影响术后结果的重要因素。该提案的目标是使用卷积 神经网络，也称为深度学习，用于映射手术切除组织的位置和范围 手术后的核磁共振扫描这项技术还将用于自动标记大脑中幸免于难的区域 在手术过程中。这些计算工具将使研究人员能够开发改进的方法， 预测术后健康状况 我们将通过训练卷积神经网络来开发自动化方法，以识别MRI上的大脑区域 这是在纽约大学朗格尼医学中心癫痫手术后获得的扫描结果。CNN已经 专门设计用于识别图像中的复杂空间模式， 手术后大脑变化的鉴定。计算机硬件和 分析方法的改进意味着CNN现在可以应用于高分辨率三维MRI扫描。这 项目将利用这些最新的发展，在计算图像分析，以提高我们的能力，预测 癫痫手术后的结果，因此有助于改善癫痫患者的治疗。

项目成果

High resolution automated labeling of the hippocampus and amygdala using a 3D convolutional neural network trained on whole brain 700 μm isotropic 7T MP2RAGE MRI.

• DOI: 10.1002/hbm.25348
• 发表时间: 2021-05
• 期刊: Human brain mapping
• 影响因子: 4.8
• 作者: Pardoe HR;Antony AR;Hetherington H;Bagić AI;Shepherd TM;Friedman D;Devinsky O;Pan J
• 通讯作者: Pan J