Multi-modal machine learning detection and tracking of traumatic brain injury neurodegeneration and its differentiation from Alzheimer's disease

摘要 我们的目标或SBIR第二阶段的工作是开发一种诊断工具，使用脑成像和其他生物标志物来识别 慢性创伤性脑病(CTE)和活体个体的早期阶段，并与 阿尔茨海默病(AD)和其他痴呆症。CTE是一种毁灭性的神经退行性疾病，发现于 经历过反复头部撞击(RHI)，导致认知障碍症状，导致痴呆症和情绪 以及可能导致暴力或自杀的行为障碍。虽然CTE在退役的NFL球员中宣传最多 在足球、曲棍球、军事战斗、家庭暴力、 老年人和其他人反复跌倒，有超过3亿人面临潜在风险。目前，虽然临床上 创伤性脑病综合征(TES)的诊断已经发展到提示可能的CTE，CTE只能 在尸检中被确诊，在生活中可能被误诊为阿尔茨海默病或其他痴呆症。没有治疗方法，也没有 检测可能支持介入治疗发展的早期、进行性阶段的手段。神经成像 生物标记物及其与流体生物标记物的组合有可能通过以下方式解决CTE诊断的需求 检测组成CTE进行性级联反应的大脑连接性、体积、功能和化学成分的变化 恶化。在我们的第一阶段SBIR工作中，我们将机器学习方法应用于体积(T1)和扩散张量 克利夫兰诊所职业拳击手脑健康研究中拳击手的磁共振成像(DTI)扫描 (PFBHS)。我们展示了TES患者和可能的CTE患者的影响和分化的渐进式模式 脑萎缩对创伤性脑损伤(TBI)和AD相关认知障碍患者的影响 损害，以及与tau的初步关系。我们第二阶段的目标是将这项工作扩大到包括不同的人群 RHI，受试者内部纵向数据分析，以及纳入功能成像和流体生物标记物，以实现 广泛适用的商用工具，可以(A)检测和区分CTE和AD，以及(B)检测和分期 脑外伤的早期进展性效应。我们将使用一个独特的全面的多模式MRI，tau PET，临床数据集 终点和流体生物标记物，来自PFBHS集合中的719名拳击手、混合武术艺术家、武术艺术家和对照组， 其中165人至少有3次影像检查；(B)240名前职业和大学橄榄球运动员和对照(诊断- CTE)；(C)219名大学接触体育运动员和对照人员(CARE)；(D)600名有创伤和/或职位的越战退伍军人 创伤应激障碍和对照(ADNI-DOD)；和(E)我们超过30,000名MRI和 代表一系列认知正常和认知受损状态的个体的PET扫描 AD和其他痴呆症。在第一阶段成功的基础上，我们将开发扩展的规范变量和深度 使用成像和流体生物标记物的学习分类器，可应用于临床评估有以下病史的人 RHI。我们的专家顾问提供的有关临床实用性和可解释性的意见将用于指导报告设计。这些 旨在为支持CTE鉴别诊断和治疗的商业产品和服务提供基础 发展。

ABSTRACT The goal or our SBIR Phase II work is to develop a diagnostic tool using brain imaging and other biomarkers to identify Chronic Traumatic Encephalopathy (CTE) and preceding stages in living individuals, and to differentiate these from Alzheimer’s disease (AD) and other dementias. CTE is a devastating neurodegenerative disorder found in individuals who have experienced repetitive head impact (RHI), causing symptoms of cognitive impairment that lead to dementia, and mood and behavioral disturbances that may lead to violence or suicide. While CTE has been most publicized in retired NFL players and “punch drunk” boxers, exposure to repetitive head impact occurs in soccer, hockey, military combat, domestic violence, repeated falls in elderly, and other persons, with over 300,000,000 individuals at potential risk. Currently, although a clinical diagnosis of Traumatic Encephalopathy Syndrome (TES) has been developed to suggest probable CTE, CTE can only be diagnosed at autopsy and can be misdiagnosed during life as AD or other dementias. There are no treatments and no means to detect earlier, progressive stages that could support the development of interventional treatments. Neuroimaging biomarkers and their combination with fluid biomarkers have the potential to address the need for a CTE diagnostic by detecting changes in brain connectivity, volume, function, and chemistries that comprise CTE’s progressive, cascade-like deterioration. In our Phase I SBIR work, we applied machine learning methods to the volumetric (T1) and diffusion tensor (DTI) magnetic resonance imaging (MRI) scans of fighters in the Cleveland Clinic Professional Fighters Brain Health Study (PFBHS). We demonstrated a progressive pattern of effects and differentiation of persons with TES and likely CTE, patterns of atrophy differentiating the effects of traumatic brain injury (TBI) from those in patients with AD related cognitive impairment, and preliminary relationships to tau. Our Phase II Aims expand this work to include different populations with RHI, within-subject longitudinal data analyses, and inclusion of functional imaging and fluid biomarkers toward achieving a broadly applicable commercially available tool that can (a) detect and differentiate CTE from AD and (b) detect and stage earlier progressive effects of TBI. We will use a uniquely comprehensive data set of multi-modality MRI, tau PET, clinical endpoints, and fluid biomarkers from (a) 719 boxers, mixed martial artists, martial artists, and controls in the PFBHS set, of whom 165 have at least 3 imaging visits; (b) 240 former professional and college football players and controls (DIAGNOSE- CTE); (c) 219 collegiate contact sports athletes and controls (CARE); (d) 600 Vietnam veterans with TBI and/or Post Traumatic Stress Disorder and controls (ADNI-DOD); and (e) individuals from our reference set of over 30,000 MRI and PET scans from individuals representing a spectrum of cognitively normal and cognitively impaired states associated with AD and other dementias. Building on our success from Phase I, we will develop expanded Canonical Variate and deep learning classifiers using imaging and fluid biomarkers that can be applied in the clinic to evaluate persons with a history of RHI. Input regarding clinical utility and interpretability from our expert Advisors will be used to guide report design. These Aims provide the foundation for commercial products and services supporting CTE differential diagnosis and treatment development.