CRCNS: Dimensionality Reduction in Cortico-Muscular Control of the Hand

描述（由申请人提供）：当你拿着咖啡杯时，你的运动皮层中的数千个神经元控制着大约40块肌肉的活动，这些肌肉控制着你手的22个骨骼自由度。控制这种日常行为的复杂性似乎令人生畏。但最近的研究表明，由于手部许多骨骼自由度的运动是高度相关的，因此在22个自由度中，多达90%的运动可以被捕获在2到7个主成分中。换句话说，描述手的大部分运动所需的维度数量可以从22个减少到7个或更少。同样，其他同时记录19块肌肉的肌电活动的研究表明，多达80%的同时记录的肌电活动可以表示为3到5个随时间变化的肌肉协同作用。因此，描述肌肉活动所需的维度数量可以从19个减少到5个或更少。这种维数的减少是否会简化控制这些日常运动的复杂性？在这里，我们建议测试一般的假设，即手部和手指的皮质肌肉控制利用降维。通过在同时记录的三个不同层次（神经、肌肉和运动）上降维，我们将采用一种新颖、全面的方法来比较所有三个层次上降维空间之间的对应关系。通过这些比较，我们将探索以前未经检验的假设：1)特定手指肌肉的生物力学结构产生手和手指运动学的某些主要成分；2)时变肌肉协同作用对应手和手指运动学的主成分；3)时变神经元协同效应代表手和手指运动学的主成分；时变神经元协同作用代表时变肌肉协同作用。为了验证我们的假设，我们将同时从植入初级运动皮层手部表征的128个单个神经元微电极、植入不同肌肉的16个肌电图电极和23个追踪手指运动的标记中获取数据，这些数据来自16到48个不同物体的抓取运动。利用这些数据，我们将提取时变神经元协同效应、时变肌肉协同效应以及手和手指运动学的主要成分。我们将确定单个肌肉、时变肌肉协同作用和/或神经元协同作用是否对应手部运动学的主要成分，以及时变神经元协同作用是否对应肌肉协同作用。如果不同层次（神经、肌肉和运动）的降维空间不能对应，我们的假设将被拒绝。相反，在不同的降维空间中，元素之间的紧密关系将支持手和手指的皮质-肌肉控制实际上利用降维的概念。

DESCRIPTION (provided by applicant): As you grasp your coffee cup, thousands of neurons in your motor cortex control the activity of some 40 muscles that move your hand's 22 skeletal degrees of freedom. The complexity of controlling such an everyday action seems daunting. But recent studies have shown that because the movements of many skeletal degrees of freedom in the hand are highly correlated, as much as 90% of the motion of the 22 degrees of freedom can be captured in only 2 to 7 principal components. In other words, the number of dimensions needed to describe most of the motion of the hand can be reduced from 22 down to 7 or fewer. Similarly, other studies in which electromyographic activity has been recorded simultaneously from 19 muscles have shown that up to 80% of the simultaneously recorded electromyographic activity can be expressed as 3 to 5 time-varying muscle synergies. The number of dimensions needed to describe muscle activity thereby can be reduced from 19 down to 5 or fewer. Might such dimensionality reduction simplify the complexity of controlling such everyday movements? Here we propose to test the general hypothesis that cortico-muscular control of the hand and fingers makes use of dimensionality reduction. By reducing dimensions at three different levels of simultaneously recorded data-neuronal, muscular and kinematic-we will take the novel, comprehensive approach of comparing the correspondence between the reduced spaces at all three levels. Through these comparisons, we will explore the previously unexamined hypotheses that: 1) the biomechanical structure of particular finger muscles produces certain principal components of hand and finger kinematics; 2) time-varying muscle synergies correspond to principal components of hand and finger kinematics; 3) time-varying neuron synergies represent principal components of hand and finger kinematics; and 4) time-varying neuron synergies represent time-varying muscle synergies. To test our hypotheses, we will acquire data simultaneously from 128 single neuron microelectrodes implanted in the primary motor cortex hand representation, from 16 electromyographic electrodes implanted in various muscles, and from 23 markers tracking finger kinematics, during grasping movements of 16 to 48 different objects. Using these data, we will extract time-varying neuron synergies, time-varying muscle synergies, and principle components of hand and finger kinematics. We will determine whether individual muscles, time-varying muscle synergies, and/or neuron synergies correspond to principal components of hand kinematics, and whether time-varying neuron synergies correspond to muscle synergies. Our hypotheses will be rejected if the spaces of reduced dimensionality at different levels-neuronal, muscular and kinematic-fail to correspond. In contrast, strong relationships between elements in the different reduced spaces would support the notion that cortico-muscular control of the hand and fingers actually utilizes dimensionality reduction. In addition to the long term benefit to society of an improved understanding of how the brain controls movement, the proposed project will have ramifications in the growing field of neuroprosthetics. Dimensionality reduction in the cortico-muscular system would provide a means of minimizing the on-line computational load carried by on-board computers that will control neurally driven prosthetic devices. More broadly, our approach may provide a model for computational reduction and interpretation of large, complex, behavioral and cognitive neuroscience datasets. Our proposal builds upon a relatively new collaboration between Schieber at the University of Rochester, who brings expertise in motor systems physiology, and Thakor at Johns Hopkins University, who brings expertise in biomedical engineering approaches to computation. Through frequent videoteleconferencing and 2-3 month exchange visits, these two labs will provide cross-disciplinary training for the co-PIs, graduate students, and undergraduates (including under-represented minorities) at both institutions. Biomedical engineers from Hopkins will learn to record physiological data while at Rochester. Motor physiologists from Rochester will learn advanced mathematical techniques for analysis while at Hopkins. The students from both groups will present their work at both neuroscience and engineering conferences, where the co-PIs will organize hands-on workshops for further dissemination of the findings per se, and of the project as a model for inter-disciplinary research. The co-PIs also will coordinate an innovative inter-institutional graduate level course.