基于泛Laguerre-Volterra模型的海马体认知神经假肢系统研究

Alzheimer’s disease is a typical neurodegenerative disease. It can affect the normal cognitive functions, such as learning and memory, through impairment to the hippocampal subregions of the patients. Thus far, there is no cure for Alzheimer’s disease with traditional medicine. In this project, we aim at designing a whole new type of hippocampal cognitive prosthetic system, based on the latest interdisciplinary research findings in electronic engineering and biological technology. One core component of our system is an implantable neural prosthetic chip, which can be used to replace the damaged subregions of the hippocampus. With the aid of this prosthetic chip, the upstream region and the downstream region of the impaired brain tissues can be bridged and normal neural signal transmission process can be reproduced. To achieve this goal, we will firstly build an enhanced nonlinear dynamical hippocampal CA3-CA1 neural network model, which can be applicable to the research of brain dynamics of non-human primates. Secondly, we will employ the heterogeneous computing platform to establish high-performance computational architecture for the generalized Laguerre-Volterra model. And finally, key functioning components of our computing system will be tested via data gathered during behavioral tasks of non-human primates, i.e. the rhesus monkey. This project will be a key stage and of good significance to the final implementation of a clinical viable cognitive neural prosthesis which can bring tremendous benefit to patients affected by neurodegenerative diseases such as the Alzheimer’s disease.

阿茨海默氏症作为典型的神经退行性疾病，可通过损伤大脑海马体子神经区域，严重影响患者的学习、记忆等高级认知功能。在药理学层面，医学界对彻底治愈阿氏症尚无有效对策。本项目立足当今电子科学和生物技术发展的新趋势，通过研发大脑海马体认知神经假肢系统，力求为阿氏症带来全新的治疗途径。该系统核心部件之一的植入式海马体神经信号编解码芯片可用以替代受损的子神经区域，完成信号在海马体内部的正常传导，进而重建大脑的学习、记忆等高级认知功能。在理论层面，本课题将研究和完善描述海马体CA3-CA1区神经信号传导的非线性动力学模型；在系统设计层面，将基于异构平台设计和研发针对CA3-CA1神经网络模型的高性能计算架构；在实验验证层面，将通过非人灵长类动物的行为学实验数据对系统进行功能验证。本项目研究工作是开发具有临床应用前景的认知神经假肢系统的重要环节，对人类神经退行性疾病的治疗和康复工程具有重要意义。

本项目从理论层面深入研究了表征大脑海马体CA3-CA1区神经尖峰信号传导的非线性动力学模型和高阶多输入态下Laguerre-Volterra模型的复杂度控制方法；从电子架构层面开发和设计了适用于Laguerre-Volterra模型关键计算模块的高精度、低复杂度、低功耗电子电路架构；从系统集成层面开发了基于现场可编程门阵列和专用集成电路平台的海马体CA3-CA1区神经尖峰信号处理系统。实验证明，相比传统计算方法，项目开发的高性能计算架构在面积、功耗和存储效能等重要指标取得了显著提升。在Laguerre-Volterra模型高斯误差单元中，其现场可编程门阵列slice LUTs, occupied slices, 以及DSP48E1s等几类资源相对同类计算方法大大降低，降低幅度可高达94.2%, 94.1%, and 90.0%。海马体CA1区神经尖峰预测单元核心面积仅为0.122 mm^2，核心动态和静态功耗有效控制在uW量级，相比软件平台归一化均方误差控制在3.67*10^(-12)，达到了符合立项预期的工程设计效果。本项目的主体研究工作是未来持续开发可用于人脑海马体神经活动编解码活动的认知神经假肢系统的重要环节与关键阶段。通过该神经假肢系统的作用，可为因海马体损伤引致的学习和记忆功能障碍等提供重要的现代临床治疗途径，因此具有重要的生物医学价值和深远的社会意义。