Explainable Machine Learning to Guide Prefrontal Brain Stimulation

项目摘要 脑刺激已经显示出对广泛的神经和精神疾病的巨大治疗前景。在 除了先进的工程工具，成功实施脑刺激需要一个全面的联合国， 了解这种治疗如何在大规模和跨网络的情况下推动网络动态和连通性的变化 多个脑区它还需要设计控制器，可以将刺激效果与行为和功能联系起来。 为了实现这些目标，我们将为精神病脑刺激开发新的可解释的机器学习模型。 为此，我们提出了三个总体目标。首先，我们的目标是学习生物学上合理和灵活的功能， 来自皮层电图（ECoG）数据的连接模型。然后，我们计划开发一个基于 在深度图卷积网络上学习ECoG数据和网络规模连接之间的关联。我们将 然后设计一个基于机器学习的精神病脑刺激指南。最后，我们将使用我们的工具来下- 网络是如何随着时间的推移而发展的。为了实现这些目标，该项目汇集了一个跨学科 在阿尔蒂官方智能和机器学习，计算和理论方面拥有独特专业知识的研究人员团队 神经科学，网络科学和生物统计学，生物工程和脑刺激实验，以及介入 精神病学和神经工程学该团队将领导实验和计算工作，将产生广告- 先进的可解释的机器学习解决方案，通过大脑刺激实验获得信息，并利用这些工具来 设计更有效的脑刺激疗法。

Project Summary Brain stimulation has shown great therapeutic promise for a wide range of neurological and psychiatric disorders. In addition to advanced engineering tools, successful implementation of brain stimulation requires a comprehensive un- derstanding of how this treatment drives changes in network dynamics and connectivity at a large scale and across multiple brain areas. It also requires the design of controllers that can relate stimulation effects to behavior and function. To achieve these goals, we will develop novel explainable machine learning models for psychiatric brain stimulation. To do so, we put forward three overarching goals. First, we aim to learn biologically plausible and flexible functional connectivity models from electrocorticography (ECoG) data. Then, we plan to develop a computational model based on a deep graph convolutional net to learn associations between ECoG data and network-scale connectivity. We will then design a machine learning based guide for psychiatric brain stimulation. Finally, we will use our tools to under- stand how the network evolves through time. To achieve these goals, the project brings together an interdisciplinary team of investigators with unique expertise in artificial intelligence and machine learning, computational and theoretical neuroscience, network science and biostatistics, bioengineering and brain stimulation experiments, and interventional psychiatric and neural engineering. The team will lead experimental and computational efforts that will produce ad- vanced explainable machine learning solutions informed by brain stimulation experiments and utilize these tools to design more efficient and effective brain stimulation therapies.