Disentangling the anatomical, functional and clinical heterogeneity of major depression, using machine learning methods

摘要 神经精神障碍的特征是不同的和共同的临床特征，出现在 不同的症状特征。描述临床症状的神经生物学病因学一直是 这是20多年来神经精神病学研究的主要目标。从最早的研究开始， 神经成像研究发现了大脑局部结构和功能的异常。然而，我们 要知道，神经精神障碍患者的大脑结构和功能存在显著的异质性。 我们团队开发的成像分析和机器学习方法提供了分析方法 需要对神经精神障碍的异质性进行量化。在这里，我们利用这些方法， 与广泛的国际合作，提供了一个独特的资源，大型，高度表型 数据集，以量化严重抑郁障碍(MDD)的异质性。在拟议的项目中， 我们专注于MDD的神经解剖学和神经功能连接性。我们的目标是识别成像特征 在MDD中应用最先进的协调、模式分析和机器学习 方法对结构和静息状态的功能磁共振成像。分析方法使我们能够量化 构成MDD的神经解剖学和神经功能连接模式，提供强大的 个人层面的预测性标记。我们的目标是达到一种新的神经解剖-神经功能 (nA-nF)MDD中的维度坐标系(MDD坐标)，其中每个维度反映 一种不同的大脑变化模式，因此捕捉到了可量化的、 可复制的、基于神经生物学的指标。我们将利用我们汇集的队列中的数据，包括 4973名成人首次发作并复发MDD，目前处于抑郁发作，这不是治疗 耐药，全部不用药，以及各自的健康对照。聚集了这些庞大而强大的 数据集将允许我们测试我们的第一个假设，即神经解剖学和神经功能 表型将表现出高度的异质性，这将使我们能够定义病理的NA-NF维度。 然后，我们检验了第二个假设，即这种异质性将与疾病相关 MDD的表型和不同的临床结局模式。我们的具体目标将得到细化和应用 先进的协调方法，以便建设性地汇集和整合这一独特的资源；2) 解剖MDD神经解剖和功能的异质性，从而得出神经成像- 基于坐标系；3)将这些成像维度与临床表型测量相关联，包括 对治疗的反应。

Abstract Neuropsychiatric disorders are characterized by distinct as well as shared clinical features that present in heterogeneous symptom profiles. Delineating the neurobiological etiology of clinical symptoms has been a key overarching aim of over 20 years of neuropsychiatric research. From the earliest studies, neuroimaging research has identified abnormalities in regional brain structure and function. However, we know that there is significant heterogeneity in brain structure and function in neuropsychiatric disorders. Imaging analytic and machine learning methods developed by our group provide the analytic approaches needed to quantify heterogeneity in neuropsychiatric disorders. Herein we leverage these methods, along with a broad international collaboration which provides a unique resource of large highly phenotyped datasets, in order to quantify heterogeneity in major depressive disorder (MDD). In the proposed project, we focus on neuroanatomy and neurofunctional connectivity in MDD. We aim to identify imaging signatures and subtypes in MDD by applying state-of-the-art harmonization, pattern analysis and machine learning methods to structural and resting state functional MRI. The analytic methods allow us to quantify the neuroanatomical and neurofunctional connectivity patterns that comprise MDD to provide powerful predictive markers at the individual level. Our goal is to arrive at a new neuroanatomical-neurofunctional (NA-NF) dimensional coordinate system in MDD (MDD COORDINATES), whereby each dimension reflects a different pattern of brain alterations, hence capturing the underlying NA-NF heterogeneity in quantifiable, replicable, and neurobiologically-based metrics. We will leverage data from our pooled cohorts consists of 4,973 adults with first episode and recurrent MDD, in a current depressive episode, that is not treatment resistant, all medication-free, and respective healthy controls. Assembling these large and powerful datasets will allow us to test our first hypothesis, namely that neuroanatomical and neurofunctional phenotypes will display high heterogeneity, which will allow us to define NA-NF dimensions of pathology. We then test the second hypothesis, namely that this heterogeneity will relate to disease-related phenotypes in MDD and different patterns of clinical outcome. Our specific aims will 1) refine and apply advanced harmonization methods in order to constructively pool and integrate this unique resource; 2) dissect the heterogeneity of the neuroanatomy and function in MDD, thereby deriving a neuroimaging- based coordinate system; 3) relate these imaging dimensions with clinical phenotypic measures, including response to treatment.