CRII: III: A Spatio-Temporal Data Mining Framework For Functional Neuroimaging Data

人类的大脑是一个由数十亿神经元组成的相互连接的网络，它使人类能够记忆、推理、感知、想象和行动。了解大脑中神经元活动与其功能之间的关系，对于精神疾病的表征、早期诊断和有效治疗至关重要。脑成像技术的进步使研究人员能够在受试者休息或工作时收集大量的大脑活动数据。多个这样的公开可用的脑成像数据集为研究大脑活动和大脑功能之间的关系提供了巨大的机会。然而，限制进展的一个主要因素是缺乏合适的计算数据挖掘工具，这些工具可以筛选具有挑战性属性的大量数据，以发现有关大脑功能的见解。开发必要工具的一个主要挑战是由于大脑活动数据的特性不同于大多数计算工具最初开发的传统研究数据。另一个挑战是由于在大脑活动数据中所表现的理想见解的方式。这个项目将产生新的计算工具和技术来解决这两个普遍的挑战。这项工作有望加速精神疾病有效治疗程序的进展。该项目的总体目标将通过定义原始神经成像数据分析问题，而不是将其硬塞到现有框架中，解决神经成像数据独特的时空特征，并利用神经科学领域的知识来完成。驱动神经科学的问题包括：1)依附于大脑连接的底层结构的功能活动的表征是什么？2)哪些脑活动图可以用来表示各种脑功能并研究它们之间的关系？3)与静态表征相比，短暂大脑状态在唯一识别对象方面的效用是什么？相应的计算研究涉及开发以下技术：1)确定大脑包裹，使所得包裹反映底层地形连通性；2)同时学习多任务- fmri数据集的字典和分类模型；3)基于fMRI数据发现并利用脑瞬态状态及其转换来唯一识别受试者。由此产生的工具和技术将使研究与个性化神经科学相关的假设成为可能——理解个体受试者共享和独特的神经过程。这将有助于实现个性化神经科学的临床相关目标，并最终减轻精神疾病的巨大社会负担。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The human brain is an interconnected web of billions of neurons that enables humans to memorize, reason, perceive, imagine, and act. Understanding the relation between neuronal activity in the brain and the functionality it enables is crucial to the characterization, early diagnosis, and effective treatment of mental illness. Advances in brain imaging technologies allow researchers to collect large volumes of brain activity data while subjects are resting or while working on a task. Multiple such brain imaging datasets that are publicly available present a tremendous opportunity to study the relationship between brain activity and brain functionality. However, a major factor limiting progress is the lack of suitable computational data mining tools that can sift through large volumes of data with challenging properties to discover insights about brain functionality. One major challenge in developing the necessary tools is due to the properties of the brain activity data that are different from traditionally studied data for which majority of the computational tools are originally developed. Another challenge is due to the manner in which desired insights are represented in the brain activity data. This project will result in novel computational tools and techniques that will address these two general challenges. This work is expected to accelerate progress towards effective treatment procedures for mental illness.The overall goals of this project will be accomplished by defining original neuroimaging data analytic problems without shoe-horning them into existing frameworks, tackling the unique spatio-temporal characteristics of neuroimaging data, and leveraging domain knowledge in neuroscience. The driving neuroscience questions include: 1) What are the representations of the functional activity that adheres to the underlying structure of the brain connections? 2) What are the brain activation maps that can be used to represent a variety of brain functions and to study relationships among them? 3) What is the utility of transient brain states in uniquely identifying subjects, in comparison to a static representation? The corresponding computational research involves developing techniques for: 1) Determining the brain parcellation such that the resultant parcels reflect the underlying topographic connectivity; 2) Simultaneously learning the dictionary as well as classification models for multiple task-fMRI datasets; 3) Discovering and using transient brain states and their transitions to uniquely identify subjects based on their fMRI data. The resultant tools and techniques will enable the investigation of hypotheses relevant to personalized neuroscience -- understanding the neurological processes that are shared and unique to individual subjects. This will help achieve the clinically relevant goals of personalized neuroscience and eventually alleviate the huge societal burden of mental illness.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.