Diagnostic Machine Learning Algorithm to Identify MEG Features of Mild TBI and Comorbid PTSD

轻度创伤性脑损伤（mTBI）是持续的身体、认知、情感和精神损伤的主要原因。 OEF/OIF/OND退伍军人和公众的行为缺陷。然而，潜在的病理生理学 和恢复机制，特别是那些与mTBI中的认知功能相关的机制， 明白mTBI后PTSD风险增加的神经元机制甚至更不清楚。 即使在持续性脑震荡后的患者中，常规MRI和CT图像通常也是阴性的。 症状（PCS）和/或PTSD症状。已经开发了基于扩散的MRI技术来识别 由于弥漫性轴索损伤（DAI）在mTBI中的主要作用，白质束异常。还甚至 复杂的基于扩散的MRI技术对于可靠的临床应用来说不够灵敏。 最近的动物研究表明，灰质也容易受到DAI的影响，这会导致异常的 从受伤区域发出的电磁信号。在这方面，支持的敏感性越来越高， 用于检测神经元异常的静息态脑磁图（rs-MEG）源成像标记 在mTBI中。我们证明，rs-MEG δ波（1-4 Hz）标记物在区分 来自神经系统完整个体的持续PCS的mTBI患者。我们还发现rs-MEG γ- 频带（30-80 Hz）标记物显示mTBI中明显的活动过度，可能是由于GABA能神经元损伤所致。 小清蛋白阳性（PV+）中间神经元。此外，我们发现，任务诱发脑磁图（te-MEG）记录， 在工作记忆（WM）任务中，检测到mTBI中整个大脑的异常信号，这些信号与 认知功能更差本申请的一个主要目标是开发高度敏感的诊断算法 将mTBI退伍军人与mTBI和PTSD共病者以及健康对照者区分开来， 老兵新方法将使用基于人工神经网络的机器学习技术， 集成rs-MEG和te-MEG成像标记器。我们将研究三组退伍军人（每组N=75）：1） 患有mTBI和持续性PCS的个体（仅mTBI组）; 2）患有mTBI和PTSD共病的个体 有持续PCS和PTSD症状的患者; 3）健康对照（HC）。目标1将建立一个机器- 用于mTBI的基于学习的MEG诊断算法，其最佳地整合了三个MEG区域成像 标记（即，δ-波段和γ-波段rs-MEG; WM诱发的MEG）以区分患有mTBI的退伍军人 （仅mTBI和共病mTBI-PTSD），准确率>90%。我们预测敏感的 mTBI分类的特征将包括rs-MEG δ和γ带活动的异常增加， 仅mTBI组和共病组的前额叶和后顶叶区以及异常WM诱发活动 相对于HC组。目标2将开发一种机器学习MEG算法，该算法集成了rs-MEG活动 和te-MEG反应诱发的负面情绪加工图片（NEPP）任务，以区分退伍军人 与mTBI-PTSD共病患者的mTBI-only比较，准确率> 90%。我们预测共病mTBI- PTSD组将显示杏仁核的rs-MEG（β带）和NEPP te-MEG活性增加， 腹内侧前额叶皮质（vmPFC）、背外侧PFC（dlPFC）和楔前叶的活动减少 mTBI组。目的3将检查异常的基于MEG的神经生理学的相关性， 仅mTBI和mTBI-PTSD共病患者的临床症状、认知障碍和现实世界的特征 生活质量我们预测，rs-MEG和WM te-MEG在特定的前额叶和后顶叶区域将 与PCS症状和认知缺陷相关。在mTBI-PTSD共病组中，PTSD症状将 与杏仁核中rs-MEG和NEPP te-MEG过度活跃以及rs-MEG异常相关 VMPFC、DLPFC和楔前叶活动减退。该项目的成功将大大提高 基于神经成像的技术，可以有效地帮助诊断mTBI，并更好地表征 mTBI和PTSD的神经生物学、神经心理学和神经精神学效应之间的关系。

Mild traumatic brain injury (mTBI) is a leading cause of sustained physical, cognitive, emotional, and behavioral deficits in OEF/OIF/OND Veterans and the general public. However, the underlying pathophysiology and recovery mechanisms, especially those associated with cognitive functioning in mTBI, are not completely understood. The neuronal mechanisms for the increased risk of PTSD after an mTBI are even less clear. Conventional MRI and CT images are generally negative even in patients with persistent post-concussive symptoms (PCS) and/or PTSD symptoms. Diffusion-based MRI techniques have been developed to identify abnormalities in white-matter tracts, owing to the major role of diffuse axonal injury (DAI) in mTBI. Yet even sophisticated diffusion-based MRI techniques are not sufficiently sensitive for reliable clinical applications. Recent animal studies indicate that gray matter is also vulnerable to DAI, which leads to abnormal electromagnetic signals from the injured regions. In this regard, support is mounting for the sensitivity of resting-state magnetoencephalography (rs-MEG) source imaging markers for detecting neuronal abnormalities in mTBI. We demonstrated that rs-MEG delta-wave (1-4 Hz) markers were very sensitive in distinguishing mTBI patients with persistent PCS from neurologically intact individuals. We also found that rs-MEG gamma- band (30-80 Hz) markers show marked hyperactivity in mTBI, possibly due to injury of GABA-ergic parvalbumin-positive (PV+) interneurons. In addition, we found that task-evoked MEG (te-MEG) recordings during working memory (WM) task detected abnormal signals throughout the brain in mTBI that were related to poorer cognitive functioning. A main goal of this application is to develop highly sensitive diagnostic algorithms to differentiate Veterans with mTBI from those with comorbid mTBI and PTSD, and those healthy control Veterans. The new approaches will use artificial neural network based machine-learning techniques to integrate rs-MEG and te-MEG imaging makers. We will study three groups of Veterans (N=75 per group): 1) individuals with mTBI and persistent PCS (mTBI-only group); 2) individuals with comorbid mTBI and PTSD who have persistent PCS and PTSD symptoms; 3) healthy controls (HC). Aim 1 will establish a machine- learning based MEG diagnostic algorithm for mTBI that optimally integrates three MEG regional imaging markers (i.e., delta-band and gamma-band rs-MEG; WM evoked MEG) to differentiate Veterans with mTBI (mTBI-only and comorbid mTBI-PTSD) from HC Veterans with >90% accuracy. We predict that sensitive features for mTBI classification will include abnormal increases in rs-MEG delta- and gamma-band activity in prefrontal and posterior-parietal areas and aberrant WM evoked activity in the mTBI-only and comorbid groups relative to the HC group. Aim 2 will develop a machine-learning MEG algorithm that integrates rs-MEG activity and te-MEG responses evoked by a negative emotion processing picture (NEPP) task to differentiate Veterans with mTBI-only from those with comorbid mTBI-PTSD with > 90% accuracy. We predict that comorbid mTBI- PTSD group will show increases in rs-MEG (beta-band) and NEPP te-MEG activity from amygdala and decreases in activity from ventromedial prefrontal cortex (vmPFC), dorsolateral PFC (dlPFC), and precuneus over the mTBI-only group. Aim 3 will examine the correlates of abnormal MEG-based neurophysiological features in mTBI-only and comorbid mTBI-PTSD with clinical symptoms, cognitive impairments, and real-world quality of life. We predict that rs-MEG and WM te-MEG in specific prefrontal and posterior parietal areas will correlate with PCS symptoms and cognitive deficits. In the comorbid mTBI-PTSD group, PTSD symptoms will correlate with abnormal rs-MEG and NEPP te-MEG hyperactivity in the amygdala, and abnormal rs-MEG hypoactivity in the vmPFC, dlPFC, and precuneus. The success of this project will significantly improve neuroimaging-based techniques that can effectively aid in the diagnosis of mTBI and better characterize the relationships among neurobiological, neuropsychological, and neuropsychiatric effects of mTBI and PTSD.