CRCNS: Computation-Enabled Adaptive Ventilatory Control System

描述(申请人提供)：大约27万名美国人和20000名法国人是创伤性脊髓损伤(SCI)的幸存者，每年有12000名美国人和1200至2000名法国人在新的损伤中存活。颈髓是最常见的损伤部位(54%)，颈椎脊髓损伤患者可能部分或完全失去呼吸控制。大多数需要呼吸机管理的脊髓损伤患者最初接受正压机械呼吸机支持，这会导致严重的不适、横隔膜萎缩、肺不张和气压创伤，并可能导致呼吸系统疾病，阻碍最佳康复。或者，也可以通过电刺激膈神经进行隔膜起搏来实现通风。最近，多块呼吸肌的肌肉内刺激被认为是一种可行的、侵入性较小的方法，特别是对于仅通过隔膜起搏呼吸不足的人。目前用于起搏的开环刺激策略有很大的局限性，包括需要手动调整刺激参数，以及无法改变肌肉疲劳或改变代谢需求的刺激参数。该方案的智能优点在于设计、开发和原型实现了一种新型的闭环控制系统，该系统利用基于脉冲的神经形态硬件的计算能力自适应地控制生物系统中的动态过程。它将具体解决同时调整振荡驱动的节奏和模式以实现对复杂生物功能的有效和高效控制的挑战。这项工作将集中在通过电刺激驱动呼吸肌肉系统的运动神经元来控制高水平脊髓损伤患者的通风的具体问题。这个问题特别适合于评估我们的方法，因为它提出了一个挑战，即在非常短的时间尺度(尖峰频率)向一组执行器提供刺激模式，以驱动协调行动，这些行动决定了只能在更长的时间尺度(呼吸频率)上评估的生理结果。拟议的计算型自适应呼吸机控制系统(CENAVEX)的开发将受益于美国(Jung)和法国(Renaud)Co-Pis及其研究团队先前的经验。美国团队在实施模式生成器/模式整形器自适应控制策略方面拥有丰富的研究经验，该策略具有在线学习功能，用于对人和啮齿动物不完全或完全截瘫后肢体运动的功能性电刺激进行计算机控制。法国团队在开发模拟和混合神经形态VLSI以及在连接活神经元和人造神经元的混合系统上的尖峰神经网络实时硬件仿真平台方面拥有丰富的研究经验。为了实现我们的目标，我们将开发一个肺-呼吸肌计算模型并测试CENAVEX系统的性能，在软件中实现用于麻醉完整啮齿动物和慢性颈椎不完全脊髓损伤患者的实时计算机控制呼吸的控制方案，并在神经形态硬件中实现该方案，包括棘波网络、突触学习和用于啮齿动物独立系统评估的生物接口硬件。这个项目的更广泛的影响在于产生了一些策略和神经形态设计，这些设计可能对一些问题有用，在这些问题中，需要在短时间尺度上协调一组执行器的振荡节奏和模式，以控制复杂的过程，在更长的时间尺度上具有动力学。拟议项目的成功完成将为转化为创新的呼吸起搏系统铺平道路，该系统能够在考虑到肌肉激活、肌肉疲劳和个人新陈代谢需求的非线性特性的情况下，允许有呼吸控制受损的脊髓损伤患者进行足够的通风。它还将为临床医生和护理人员提供轻松的部署。通过提供长时间的呼吸训练，该系统可以作为不完全性脊髓损伤患者的康复工具，提高使用者的生活质量。多学科研究工作将通过国际人员和思想交流，在学科和国际机构之间架起桥梁。佛罗里达国际大学是一所少数民族和西班牙裔服务机构，将提供接触多样化学生群体的机会，该项目将直接支持女性博士后、年轻研究员、研究生和本科生的培训。培训部分将建立神经科学、生物力学、康复、神经形态工程和神经控制系统方面的跨学科专门知识，并将向受训者提供使用计算神经科学方法解决在开发嵌入式神经形态技术方面面临的复杂挑战的技能。

DESCRIPTION (provided by applicant): Approximately 270,000 Americans and 20,000 French are survivors of traumatic spinal cord injury (SCI), with 12,000 Americans and 1200 to 2000 French surviving new injuries each year. The cervical cord is the most common site of injury (54%) and people with cervical SCI can have partial or complete loss of ventilatory control. Most people with SCI that require ventilation management are initially supported with positive pressure-mechanical ventilation, which is associated with significant discomfort, diaphragm atrophy, atelectasis and barotrauma and can lead to respiratory diseases and prevent optimal recovery. Alternatively, ventilation can be achieved by diaphragmatic pacing by electrical phrenic nerve stimulation. More recently, intramuscular stimulation of multiple respiratory muscles has been proposed as a viable less surgically invasive approach, in particular for people with insufficient ventilation by diaphragmatic pacing alone. The open-loop stimulation strategy currently utilized for pacing has major limitations including the need for manual stimulation parameter tuning, and inability to alter stimulation parameters on muscle fatigue or changing metabolic demand. The intellectual merit of this proposal lies in the design, development and prototype realization of a novel closed-loop control system that utilizes the computational power of spike-based neuromorphic hardware to adaptively control dynamic processes in biological systems. It will specifically address the challenge of simultaneously adapting the rhythm and pattern of oscillatory drive to achieve effective and efficient control of complex biological functions. The work will focus on the specific problem of controlling ventilation in individuals with high-level SCI by electrically stimulating the motoneurons that drive respiratory musculature. This problem is particularly well-suited to assess our approach because it presents the challenge of delivering a pattern of stimuli to a set of actuators at very short timescales (spike frequencies) to drive coordinated actions that determine physiological outcomes that can only be assessed on a much longer timescale (breathing frequency). Development of the proposed computation-enabled adaptive ventilatory control system (CENAVEX) will benefit from the prior experience of the US (Jung) and French (Renaud) Co-PIs and their research teams. The US team has extensive research experience in implementation of a Pattern Generator/Pattern Shaper adaptive control strategy with on-line learning for computer control of functional electrical stimulation of limb movement after incomplete or complete paraplegia in people and rodents. The French team has extensive research experience in development of analog and mixed neuromorphic VLSI and real-time hardware simulation platforms of spiking neural networks on hybrid systems interfacing living and artificial neurons. To accomplish our objectives we will develop a lung-respiratory muscles computational model and test the abilities of the CENAVEX system, implement the control scheme in software for real-time computer-based control of ventilation in anesthetized intact rodents and those with chronic cervical incomplete SCI, and implement the scheme in neuromorphic hardware with spiking networks, synaptic learning and bio-interface hardware for standalone system assessment in rodents. The broader impact of this project lies in the production of strategies and neuromorphic designs that could be useful in a number of problems in which oscillatory rhythm and pattern across a set of actuators need to be coordinated on short timescales to control complex processes with dynamics over much longer timescales. Successful completion of the proposed project will pave the path for translation to an innovative respiratory pacing system capable of allowing adequate ventilation in people with SCI with impaired respiratory control, taking into account non-linear properties of muscle activation, muscle fatigue, and metabolic demand of the individual. It will also offer ease of deployment for the clinician and caregiver. By providing long duration respiratory exercise, the system could act as a rehabilitative tool in people with incomplete SCI improving the quality of life for the user. The multidisciplinary research effort will bridge disciplines and international institutions through international exchange of personnel and ideas. Florida International University, a minority and Hispanic serving institution, will provide access to a diverse student body and the project will directly support the training of a female postdoc, young investigator, graduate and undergraduate students. The training component will build transdisciplinary expertise in neuroscience, biomechanics, rehabilitation, neuromorphic engineering and neural control systems and will provide trainees with the skills to use computational neuroscience approaches to address complex challenges faced in developing embedded neuromorphic technology.