Collaborative Research: NCS-FO: Dynamic Brain Graph Mining

将人脑中的连接映射为网络系统，即脑图，已经成为神经科学中的一种普遍范式。在认知发展、衰老和疾病中，了解大脑的结构和功能如何随着时间的推移而变化，以提供对个体差异以及不同行为和障碍背后的机制的洞察是至关重要的。然而，传统的模型大多认为脑图是“静态的”，忽略了潜在的随时间变化的情况。这个项目旨在开发新的方法来对脑图的动态进行建模，这些方法在生成准确、可解释和公平的预测方面是健壮的。这个跨学科项目将为参与的研究人员提供独特的培训组合，研究成果将被纳入教育。研究人员将通过已建立的基准平台、新的出版物、教程以及与领域专家的合作来传播他们的发现。该项目寻求克服现有静态脑图模型的障碍，并开发实用的基础和计算工具来处理和分析从动态神经成像数据得出的复杂脑图。该项目将开发一个统一的脑图常微分方程(BrainGDE)框架，将先进的深度图形学习技术和常微分方程交织在一起，解决数据复杂性、模型可解释性、公平性和可信度以及临床转换方面的挑战。计划的研究任务将集中在：(1)单模式动态脑图挖掘，(2)多模式动态脑图挖掘，以及(3)与领域专家合作的临床调查。如果成功，这项研究将重塑生物信息学和医疗保健技术中时间数据挖掘的深度学习方法。该项目中建立的动态图形挖掘框架还将指导对具有结构的高维动态数据的感知、知识发现、推理和推理问题的研究，并将作为未来这一方向工作的通用基准。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Mapping the connections in human brains as networked systems, i.e., brain graphs, has become a pervasive paradigm in neuroscience. In cognitive development, aging, and disease, it is crucial to understand how the structures and functions of the brain change over time to provide insights into individual differences and the mechanisms underlying different behaviors and disorders. Traditional models, however, mostly treat the brain graphs as “static,” ignoring the underlying changes over time. This project aims to develop new methods for modeling the dynamics of brain graphs that are robust in generating accurate, interpretable, and fair predictions. This interdisciplinary project will provide a unique mix of training for the participating researchers, and the research findings will be incorporated into education. The investigators will disseminate their findings through an established benchmark platform, new publications, tutorials, and collaborations with domain experts.This project seeks to overcome the barriers of existing static brain graph models and develop practical foundations and computational tools for processing and analyzing complex brain graphs derived from dynamic neuroimaging data. The project will develop a unified framework of Brain Graph Ordinary Differential Equations (BrainGDE) interweaving advanced deep graph learning techniques and ordinary differential equations, addressing the challenges of data complexity, model interpretability, fairness and trustworthiness, as well as clinical transformation. Planned research tasks will focus on: (1) unimodal dynamic brain graph mining, (2) multimodal dynamic brain graph mining, and (3) clinical investigations, in collaboration with domain experts. If successful, this research will reshape deep learning approaches for temporal data mining in bioinformatics and healthcare technologies. The dynamic graph mining framework established in this project will also guide research on the problems of sensing, knowledge discovery, reasoning, and inference on high-dimensional dynamic data with structures and will serve as a universal benchmark for future work in this direction.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.