Neuropsychiatric Classification via Connectivity and Machine Learning

描述（由申请人提供）：精神疾病诊断和统计手册（DSM）中体现的神经精神疾病的诊断系统是基于症状群，而不是基于潜在的病因学或病理生理学。30年前，建立可靠的诊断是精神病学科学进步的关键一步，但现在，它掩盖了大脑生物学和个体患者症状之间的关系，从而阻碍了该领域的发展--这种关系在最好的情况下也是模糊的。这一认识促使人们以NIMH的研究领域标准（RDoC）倡议的形式寻找一种替代的、基于大脑的诊断系统。这种替代诊断框架的发展还处于起步阶段，需要新的策略对病理生理状态进行合理的分类。 我们已经成功地使用了数据驱动的功能连接数据分析，来自休息时大脑的功能神经成像。这种方法揭示了几种神经精神疾病的神经连接障碍。我们将应用这些数据驱动的方法，结合领先的机器学习算法，来量化主要DSM疾病之间和内部的连接障碍模式。 我们已经收集了707个静息状态扫描的数据集，这些扫描是在最先进的3 T扫描仪上进行的，并通过了严格的质量控制标准，包括五种主要的DSM诊断：精神分裂症，双相情感障碍，重度抑郁症，强迫症和创伤后应激障碍，每种都有匹配的对照。伴随症状评估由高技能人员进行。这个大型混合数据集允许前所未有的跨诊断，数据驱动搜索跨诊断的共享或不同的连接障碍。具体来说，我们将采用一种强大的多层分析方法，使用：完全数据驱动的连接分析，重点是通过健康受试者的工作先验定义的网络，以及基于种子的方法，重点是与组成DSM诊断相关的电路。 我们假设了几种可能的结果。首先，从数据驱动的连接性分析中得出的患者群体确实可以映射到基于DSM的DSM诊断。这将是一个以症状为中心的疾病分类学的验证，至少在这些条件下。第二，数据驱动的分析可以识别贯穿DSM诊断的新类别。第三，结果可能遵循连续的连接障碍，如RDoC框架提出的那些。一个更复杂的结果，混合这些模式也是可能的。最后，紧急模式将与症状措施，内部和跨疾病。 无论最终的模式，这个项目的结果将严格通知正在进行的努力，以完善一个诊断方案，精神疾病是牢固地建立在他们的病理生理学。此外，该方法将适用于其他数据集。我们预计，这种方法将提供一个关键支柱的发展，以大脑为基础的理解异质性的精神疾病。

DESCRIPTION (provided by applicant): The diagnostic system for neuropsychiatric conditions embodied in the Diagnostic and Statistical Manual of Psychiatric Disorders (DSM) is based on clusters of symptoms rather than on underlying etiology or pathophysiology. The establishment of reliable diagnoses was a critical step in the advancement of psychiatric science three decades ago, but now it holds the field back by concealing relationships between brain biology and individual patients' symptoms - relationships that are obscure under the best of circumstances. This realization motivates a search for an alternative, brain-based diagnostic system, in the form of the NIMH's Research Domain Criteria (RDoC) initiative. The development of such an alternative diagnostic framework is in its infancy, and new strategies are needed for the rational categorization of pathophysiological states. We have successfully used data-driven analysis of functional connectivity data, derived from functional neuroimaging of the brain at rest. This approach has revealed neural dysconnectivity across several neuropsychiatric conditions. We will apply these data-driven approaches, in conjunction with leading machine learning algorithms, to quantify dysconnectivity patterns across and within major DSM disorders. We have assembled a dataset of 707 resting-state scans, performed on state-of-the-art 3T scanners and passing rigorous quality control standards, comprising five major DSM diagnoses: schizophrenia, bipolar disorder, major depressive disorder, obsessive-compulsive disorder, and post-traumatic stress disorder, with matched controls for each. Accompanying symptom assessments were administered by highly skilled personnel. This large hybrid dataset permits an unprecedented cross-diagnostic, data-driven search for shared or distinct dysconnectivity across diagnoses. Specifically, we will employ a powerful multi-tiered analytic approach using: fully data-driven connectivity analysis, focusing on networks defined a priori by work in healthy subjects, and a seed-based approach focused on circuits associated with the constituent DSM diagnoses. We hypothesize several possible outcomes. First, patient groups derived from the data-driven connectivity analyses may indeed map onto symptom-based DSM diagnoses. This would be a validation of a symptom- focused nosology, at least across these conditions. Second, data-driven analysis may identify new categories that cut across DSM diagnoses. Third, results may follow continua of dysconnectivity, such as those proposed by the RDoC framework. A more complex outcome that blends these patterns is also probable. Finally, emergent patterns will be correlated against symptom measures, within and across disorders. Irrespective of the ultimate pattern, results of this project will critically inform ongoing effort to refine a diagnostic scheme for psychiatric disorders that is firmly grounded in their pathophysiology. Furthermore, the methodology will be applicable to other datasets. We anticipate that this approach will provide a key pillar to the development of a brain-based understanding of the heterogeneity of psychiatric disease.