Dynamic modulation of brain states using brain stimulation and neuroadaptive Bayesian optimization

就像一个交响乐团依赖于其成员的协调努力一样，大脑依赖于它的许多区域一起工作，以执行使我们成为人类的多种认知功能。这些功能允许我们解决问题，从内存中检索相关信息，并选择执行特定任务所需的响应。为了做到这一点，大脑必须协调相隔很远的区域之间的相互作用。现代神经科学的最大挑战之一是理解这些相互作用是如何发生的，以及它们的发生如何导致有效的行为。一种能够影响大脑区域之间相互作用的工具可以帮助科学家更好地理解大脑活动的特定模式是如何与有效行为相关联的，例如能够在记忆中保留信息或解决问题。这样的工具可以应用于神经和精神疾病，在这些情况下，大脑区域之间的相互作用可能会出现故障。该项目的目标是开发这种工具。为了做到这一点，我们将结合功能磁共振成像(FMRI)、无创脑电刺激和机器学习。这些技术中的每一种都为这个工具带来了一个关键元素。FMRI是神经科学家广泛使用的一种技术，它为图像提供关于大脑功能的信息。无创脑电刺激是一种通过头皮施加低电压电流的技术，可以改变神经元的活动，而不需要手术植入电极。这项技术已经被证明可以影响大脑功能和大脑区域之间的相互作用。然而，大脑电刺激可以以许多不同的方式应用，因此很难知道是什么在影响一组大脑区域之间的特定相互作用。此外，大脑刺激的结果可能会因年龄、性别、大脑解剖和遗传等因素而有所不同。这使得创建一种能够识别每个人的刺激参数的工具就像大海捞针一样。这就是为什么机器学习是必要的，计算机程序在使用传统方法所不可能的时间框架内，通过机器学习来确定哪些大脑刺激参数最适合参与认知功能的大脑区域。本质上，我们的工具将使用大脑刺激来影响大脑区域如何相互作用，当参与者接受某种类型的刺激时，分析fMRI数据以告知大脑对该刺激的反应，而机器学习将选择下一个应该研究的刺激。在实验结束时，我们将获得大脑对不同刺激条件的反应图，并预测引发大脑反应的最佳刺激条件是什么。然后，这种工具可以用于许多临床条件下观察到的大脑区域之间沟通效率低下的情况，如精神疾病和脑损伤后的康复。

Like an orchestra that relies on the coordinated efforts of its members, the brain depends on its many regions working together to perform the multitude of cognitive functions that makes us human. These functions allow us to solve problems, retrieve relevant information from memory and select the responses necessary to perform a particular task. In order to do this, the brain must coordinate the interactions between regions located far apart. One of the greatest challenges of modern neuroscience is to understand how these interactions occur, and how their occurrence gives rise to efficient behaviour. A tool capable of influencing the interactions between brain regions could help scientists understand better how a particular pattern of brain activity is associated to efficient behaviour, such as being able to retain information in memory or solve a problem. Such a tool could then be applied to neurological and psychiatric conditions, where the interactions between brain regions might be malfunctioning.The objective of this project is to develop this tool. In order to do this, we will combine functional magnetic resonance imaging (fMRI), non-invasive electrical brain stimulation and machine learning. Each of these techniques brings a critical element to this tool.FMRI is a technique widely used by neuroscientists to provide images with information about brain function. Non-invasive electrical brain stimulation is a technique that applies low-voltage current through the scalp and can change the activity of neurons without requiring surgery to implant electrodes. This technique has been shown to influence brain function and the interactions between brain regions. Electrical brain stimulation, however, can be applied in many different ways, thereby making it difficult to know what would work for to influence a particular interaction between a set of brain regions. In addition, the results of brain stimulation can vary depending on factors such as a person's age, sex, brain anatomy and genetics. This makes creating a tool capable of identifying the stimulation parameters for each individual like 'finding a needle in a haystack'. This is why machine learning is necessary, where a computer program "learns" to identify which brain stimulation parameters optimally engage brain regions involved in cognitive functions in a time frame that would not be possible using conventional methodologies.In essence, our tool will use brain stimulation to influence how brain regions interact, fMRI data analysed while the participant is receiving a certain type of stimulation to inform on how the brain reacts to it, and machine learning to select the next stimulation that should be investigated. By the end of the experiment we will obtain a map with the brain's responses to different stimulation conditions, and a prediction of what the optimal stimulation condition to elicit a brain response is.This tool could then be used in many clinical conditions where inefficient communication between brain regions has been observed, such as psychiatric conditions and during rehabilitation after brain injury.