Interpretable machine learning to synergize brain age estimation and neuroimaging genetics

可解释的机器学习可协同大脑年龄估计和神经影像遗传学

• 批准号: 10568234
• 负责人: Andrei Irimia
• 金额: $ 80.69万
• 依托单位: UNIVERSITY OF SOUTHERN CALIFORNIA
• 依托单位国家: 美国
• 财政年份: 2023
• 资助国家: 美国
• 起止时间: 2023-02-01 至 2028-01-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10568234
• 关键词: 10 year old; Acceleration; Adult; Affect; Age; Aging; Alzheimer' s Disease; Alzheimer' s disease diagnosis; Alzheimer' s disease patient; Alzheimer' s disease risk; Anatomy; Architecture; Blood - brain barrier anatomy; Blood Cells; Brain; Brain region; Chronology; Clinical; Cognition; Cognitive; Communities; DNA Methylation; Data Set; Disease; Ensure; Environment; Epigenetic Process; Ethnic Origin; Exhibits; Extravasation; Genes; Genetic; Genotype; Geometry; Image; Impaired cognition; Individual; Logic; Machine Learning; Magnetic Resonance Imaging; Maps; Measures; Methods; Neuroanatomy; Neurodegenerative Disorders; Neuropsychology; Pathologic; Pattern; Persons; Phenotype; Predisposition; Process; Race; Sampling; Specificity; Techniques; Testing; Training; Validation; Whole Blood; aging brain; biobank; brain magnetic resonance imaging; cerebral microbleeds; cognitive function; cohort; dementia risk; design; genetic variant; genome wide association study; insight; methylation pattern; mild cognitive impairment; neural; neuroimaging; novel; polygenic risk score; pre-clinical; predictive modeling; resilience; sex; synergism; trustworthiness

PROJECT SUMMARY Brain age can be used a predictor of deviation from typical age trajectories due to disease processes. Because brain age is strongly associated with neurodegenerative disease, brain age predicted from magnetic resonance images (MRIs) can become an affordable and noninvasive preclinical indicator of mild cognitive impairment (MCI) and Alzheimer’s disease (AD) risk. Today’s best brain age estimation approaches use black-box machine learning (ML) that often lacks interpretability in the sense that it does not specify which neuroanatomic features are critical for brain age estimation. The first aim of this project is to design, test, and validate an interpretable ML architecture that leverages brain MRIs to estimate brain with high accuracy. We will construct an interpretable ML architecture trained on structural MRIs to identify neuroanatomic features that reflect brain age at the level of subjects and cohorts. These techniques will be tested and validated to ensure trustworthiness and generali- zability to new datasets. We hypothesize that our ML can use MRIs to predict MRI-derived brain age significantly more accurately than existing methods. Our second aim is to map neuroanatomic features that predict brain age and that reflect abnormal aging observed in MCI/AD. We will test the hypothesis that, aside from aging-related neuroanatomic features shared by cognitively normal subjects and MCI/AD patients, the latter exhibit additional neuroanatomic features that can distinguish them from the former, early during adulthood, with high sensitivity, specificity, and precision. Our third aim is to use genome-wide association (GWAS) to find genes associated with neuroanatomic features of brain aging that predict MRI-derived brain age. We will synergize our interpreta- ble ML approaches with GWAS to find genetic factors that affect brain aging features predictive of MCI/AD diag- nosis. We will develop and validate a polygenic risk score (PRS) of resilience/vulnerability to accelerated brain aging observed in MCI/AD. If successful, this project will deliver trustworthy, generalizable, and interpretable ML approaches that can leverage MRIs to identify novel brain aging features reflecting MCI/AD risk. Because aging is a lifelong process, we have the potential to detect such features much earlier than currently possible. Im- portantly, we will identify genes that act on brain aging in ways that may lead to MCI/AD. This can provide considerable insight on the potential mechanisms relating genetic factors to brain aging and MCI/AD.

项目摘要 脑年龄可以用来预测由于疾病过程而偏离典型年龄轨迹的情况。因为 脑年龄与神经退行性疾病密切相关，通过磁共振预测脑年龄 磁共振成像（MRI）可以成为轻度认知障碍的一种负担得起的非侵入性临床前指标 (MCI)和阿尔茨海默病（AD）风险。今天最好的大脑年龄估计方法使用黑盒机器 学习（ML）通常缺乏可解释性，因为它没有指定哪些神经解剖学特征 对大脑年龄的估计至关重要这个项目的第一个目标是设计、测试和验证一个可解释的 ML架构利用大脑MRI以高精度估计大脑。我们将构建一个可解释的 ML架构在结构MRI上进行训练，以识别反映大脑年龄的神经解剖学特征 受试者和队列。这些技术将被测试和验证，以确保可信度和通用性， 对新数据集的灵活性。我们假设我们的ML可以使用MRI来预测MRI衍生的大脑年龄 比现有的方法更准确。我们的第二个目标是绘制预测大脑年龄的神经解剖学特征 并且反映了在MCI/AD中观察到的异常老化。我们将检验一个假设，除了与衰老有关的 认知正常受试者和MCI/AD患者共有的神经解剖学特征，后者表现出额外的 神经解剖学特征，可以区分他们从前者，在成年早期，具有高灵敏度， 特异性和精确性。我们的第三个目标是使用全基因组关联（GWAS）来寻找相关基因， 大脑老化的神经解剖学特征来预测MRI衍生的大脑年龄。我们将协同我们的解释- ble ML方法与GWAS一起寻找影响MCI/AD诊断的脑老化特征预测的遗传因素。 感觉。我们将开发和验证一个多基因风险评分（PRS）的弹性/脆弱性，以加速大脑 在MCI/AD中观察到老化。如果成功，该项目将提供可信赖的，可推广的和可解释的ML 这些方法可以利用MRI来识别反映MCI/AD风险的新的脑老化特征。因为衰老 是一个终身的过程，我们有可能比目前可能更早地检测到这些特征。我... 更重要的是，我们将鉴定出影响大脑衰老的基因，这些基因可能导致MCI/AD。这可以提供 对遗传因素与脑老化和MCI/AD相关的潜在机制有相当多的了解。