Development of closed-loop control systems for therapeutic applications of brain-computer interfaces

该项目将探索用于治疗人类神经系统疾病的脑机接口的个性化控制算法的设计。重点将放在基于已知生物标志物的反馈控制器上，这些生物标志物与原发性震颤和帕金森病的症状有关，使用一种称为深部脑刺激的技术。目标将是创建控制器，该控制器对连接电刺激与生物标志物的动态模型中的参数不确定性和错误具有鲁棒性。控制算法还将适应不同患者之间大脑活动和结构的可变性，以及每个人因疾病进展而随时间发生的变化。使用有针对性的刺激，而不是持续的刺激，将降低这些设备的功耗，并改善患者的临床结果，特别是减少不良副作用。为了实现这一目标，该项目将考虑大脑对刺激反应的动态模型结构，确保它们能够准确预测神经行为，而不会产生过度的复杂性，这将限制模型参数拟合可行数据量的能力，并阻止其在低功耗微控制器上的使用。此外，将开发新技术来确定每个患者的模型参数，以便这些参数可用于设计鲁棒的自适应模型预测控制算法。该项目还将探索新的调度算法，根据对模型预测性能的估计，以及对已知会影响相关生物标志物（如行走、睡眠或执行到达任务）的患者活动状态的估计，确定模型更新的方式和时间。此外，该项目将开发技术，以确保参数识别数据足够丰富，并以最佳方式权衡控制器性能与模型参数信息积累率。这将提高精度，并有助于减少计算参数更新的计算成本，从而提高在嵌入式硬件上部署这种控制算法的可行性。控制算法将在硬件模拟器上实现，并与谭惠玲教授的研究小组（纳菲尔德临床神经科学系脑网络动力学组）合作在人类患者身上进行测试。该项目属于EPSRC控制工程和临床技术研究领域。

The project will explore the design of personalised control algorithms for brain-computer interfaces used to treat neurological conditions in humans. The focus will be on feedback controllers based on known biomarkers associated with essential tremor and symptoms of Parkinson's disease using a technique known as deep brain stimulation. The goal will be to create controllers that are robust to parameter uncertainties and errors in the dynamic models linking electrical stimulation to the biomarkers. The control algorithms will also adapt to the variability in brain activity and structure between different patients, as well as the variations that occur for each individual over time due to disease progression. The use of targeted, rather than constant stimulation will lead to lower power consumption of these devices, as well as improvement in clinical outcomes for patients, particularly a reduction in adverse side effects. To achieve this goal, the project will consider the structure of dynamic models of the brain's response to stimulation, ensuring that they can accurately predict neural behaviour without incurring excessive complexity, which would limit the ability to fit model parameters to feasible amounts of data and prevent its use on low-power microcontrollers. Additionally, new techniques will be developed to identify the model parameters for each patient such that these can be used to design robust adaptive model predictive control algorithms. The project will also explore new scheduling algorithms to determine how and when the model updates will occur, based on estimates of the model's predictive performance as well as on estimates of the patients' state of activity that are known to affect the relevant biomarkers, such as walking, sleeping, or performing reaching tasks.Furthermore, the project will develop techniques for ensuring that parameter identification data is sufficiently rich, and for optimally trading off controller performance against the rate of accumulation of information about model parameters. This will improve accuracy and facilitate a reduction in the computational cost of calculating parameter updates, thereby improving the feasibility of deploying such control algorithms on embedded hardware. The control algorithms will be implemented on hardware emulators and tested in human patients in collaboration with Prof Huiling Tan's research group (Brain Network Dynamics Unit at the Nuffield Department of Clinical Neurosciences). This project falls within the EPSRC Control Engineering and Clinical Technologies research areas.