Automated non-invasive brain stimulation parameter selection using Bayesian optimisation

我这个项目的总体目标是开发一种方案，利用非侵入性脑刺激有效地改变高级认知功能。非侵入性脑刺激是现代神经科学研究中的一项关键技术，它使研究人员能够探索大脑过程和行为之间的因果联系，并为患有精神疾病和神经疾病(包括帕金森氏症和阿尔茨海默氏症)的患者开发新的治疗方法。然而，这种技术的一个关键缺点是，其效果严重依赖于参与者现有的大脑状态，导致不同个体之间的刺激效果不同。在这篇博士论文中，我的目标是将神经成像和非侵入性脑刺激技术结合起来，开发出一种更有效的刺激方案，以适应参与者的大脑状态。在这项研究的主要部分，我将利用机器学习技术，根据参与者的神经状态对行为任务的试验进行分类。机器学习算法对行为试验的在线分类和由此产生的刺激管理将在闭环系统中执行，即系统一旦初始化，将能够在没有研究人员任何输入的情况下运行，任何刺激开始/参数将由算法根据参与者的大脑状态确定(在严格的安全限制内)。使用闭环设计将允许更有效的刺激方案，能够更好地适应当前的行为任务要求，从而将非侵入性脑刺激转变为更可靠的研究和治疗工具。这项工作不仅有可能通过减少脑刺激实验结果的变异性来改善研究中的脑刺激结果，而且还有可能转化为治疗环境，在那里它有可能提高临床人群的刺激效果。这项工作的完成将涉及几个MRC技能优先领域的培训，包括1)以高级统计形式的量化技能，包括混合模型分析和机器学习技术；2)通过与工程师以及心理学家和神经科学家的互动，特别是在实施机器学习技术和开发闭环系统期间的跨学科技能；3)整个器官/机体生理学，因为该项目涉及神经成像以及通过脑刺激改变脑功能的方法。

The overall aim of my project is to develop a protocol to efficiently alter high-level cognitive functions using non-invasive brain stimulation. Non-invasive brain stimulation is a key technique in modern neuroscientific research, allowing researchers to probe the causal link between brain processes and behaviour, as well as developing new therapeutic treatments for patients suffering from mental illnesses and neurological diseases including Parkinsonism and Alzheimer's. However, a key downside of this technique is that its effects are heavily reliant on the participant's existing brain state, leading to varying stimulation effects between individuals. In this PhD I will aim to combine neuroimaging and non-invasive brain stimulation techniques to develop a more efficient stimulation protocol which adapts to participant brain state. In the main part of this research I will utilise machine-learning techniques to classify trials of a behavioural task according to the participant's neural state. The online classification of behavioural trials by the machine-learning algorithm and the resulting administration of stimulation will be performed in a closed-loop system, i.e. the system, once initialised, will be able to run without any input from the researcher, and any stimulation onset/parameters will be determined by the algorithm (within strict safety limitations) depending on participant brain state. The use of a closed-loop design will allow for a more efficient stimulation protocol, which is better able to adapt to the current behavioural task demands and so transforming non-invasive brain stimulation into a more reliable research and therapeutic tool. This work has the potential to not only improve brain stimulation results in research by reducing the variability of results from brain stimulation experiments, but could also have a potential translation to therapeutic settings where it has the potential to improve the efficacy of stimulation in clinical populations. The completion of this work will involve training in several MRC skill priority areas including 1) quantitative skills in the form of high level statistics including mixed model analysis and machine learning techniques, 2) interdisciplinary skills through interaction with engineers as well as psychologists and neuroscientists, particularly during implementation of machine learning techniques and development of the closed-loop system and 3) whole organ/organism physiology as the project involves both neuroimaging as well as methods to alter brain functions using brain stimulation.