Neuroimaging and Machine Learning to Redefine Anxiety and Depression

 描述(由申请人提供)：我的目标是从大脑功能、生理和行为的多种神经生物学测量中确定一种不受现有诊断界限限制的数据驱动的抑郁和焦虑分类。焦虑障碍和严重抑郁障碍非常普遍，每年在护理和生产力损失方面的成本超过1000亿美元。虽然用于诊断这些疾病的症状传达了有用的信息并反映了真正的现象学，但症状分组的方式造成了“模糊”的诊断边界，在疾病之间有大量的症状重叠，但内部存在巨大的症状异质性。此外，旨在确定神经对功能障碍的贡献的实验本质上与这些传统诊断类别有关，因此，我们不清楚抑郁和焦虑背后的神经电路如何与生理和行为水平上的症状相关联，独立于这些传统诊断。这些模糊的诊断界限阻碍了我们在了解功能障碍的机制和开发新的、有针对性的治疗方法方面取得的进展。因此，建立焦虑和抑郁的互补特征将是有益的，它反映了不同神经原因的凝聚力簇。为了解决这些问题，我建议使用一种数据驱动的方法来为抑郁和焦虑开发一种替代分类。在目标1下，我将在600多名参与者的丰富现有数据集上使用计算方法，从情绪反应和调节的神经成像探测器中推导出情绪处理的维度结构，并确定这些结构如何与跨越行为、生理和自我报告的其他功能水平相关联。在目标2下，我将使用稀疏聚类算法根据神经成像结构对单个受试者进行分类，然后确定每个分类是如何通过行为、生理和自我报告症状测量来表达的，而不是传统的诊断。为了解决目标3，我将使用实验性的压力探头来分析神经成像和其他测量单位之间关系的状态和特征类成分。结果将是一个新的分类，将推动我们在了解抑郁症和焦虑症的神经功能障碍的机制以及开发针对此类功能障碍的新疗法方面取得进展。关键是，提出的多模式方法利用无监督机器学习算法，以一种不受当前诊断范例假设的方式识别这个复杂系统中的潜在模式。这种方法产生的特征将提供一个维度空间，以了解神经回路功能的自然变异以及这种变异如何与每个人的功能表型相关。这样的特征将是在转变对抑郁和焦虑的理解方式方面向前迈出的重要一步，消除耻辱，并允许从可以更有效地转化为临床的机械模型开发新的治疗方法。

 DESCRIPTION (provided by applicant): I aim to identify a data-driven taxonomy of depression and anxiety from multiple neurobiological measures of brain function, physiology and behavior that is not constrained by existing diagnostic boundaries. Anxiety Disorders and Major Depressive Disorder are highly prevalent and together cost over $100 billion per year in care and lost productivity. While the symptoms used in the diagnosis of these disorders convey useful information and reflect real phenomenology, the way in which symptoms are grouped makes for "fuzzy" diagnostic boundaries, with substantial symptom overlap across disorders, yet vast symptom heterogeneity within. Moreover, experiments aiming to identify the neural contribution to dysfunction have been intrinsically tied to these traditional diagnostic categories As a consequence, we do not have a clear understanding of how the neural circuitry underlying depression and anxiety relates to the expressed symptoms at the level of physiology and behavior, independent from these traditional diagnoses. These blurry diagnostic lines hamper our progress toward understanding the mechanisms of dysfunction and developing novel, targeted therapeutics. Therefore, it would be beneficial to establish a complementary characterization of anxiety and depression that reflects cohesive clusters of distinct neural causes. Addressing these issues I propose to use a data driven approach to develop an alternate classification for depression and anxiety. Under Aim 1 I will use computational methods on a rich existing dataset of over 600 participants, to derive dimensional constructs of emotion processing from neuroimaging probes of emotion reactivity and regulation and determine how these constructs are associated with other levels of function spanning behavior, physiology and self-report. Under Aim 2 I will use sparse clustering algorithms to classify individual subjects according to the neuroimaging constructs and then determine how each classification is expressed across behavioral, physiological and self-report symptom measures, independent of traditional diagnosis. To address Aim 3 I will use experimental stress probes to parse state versus trait-like components of the relationships between neuroimaging and each other unit of measurement. The outcome will be a novel classification that will advance our progress toward both understanding the mechanisms of neural dysfunction in depression and anxiety as well as developing novel therapeutics for targeting such dysfunction. Critically, the proposed multi-modal approach utilizes unsupervised machine learning algorithms to identify the underlying patterns within this complex system in a manner that is free from the assumptions of the current diagnostic paradigms. The resulting characterization from this approach will provide a dimensional space to understand the natural variation in neural circuit function and how this variation relates to each person's functional phenotype. Such a characterization will be a significant step forward in transforming the way that depression and anxiety are understood, removing stigma, and allowing novel treatments to be developed from mechanistic models that can be more effectively translated to the clinic.