Neural encoding of motor precision for advancing brain-machine interfaces

随着技术的进步，脑机接口(BMI)发展迅速，允许从大量神经元进行长期记录，更强大的计算能力，更好的训练，以及更好的回归算法来估计神经调谐。然而，目前的BMI几乎完全依赖于固定的、线性的神经编码模型，这些模型对所有类型的运动都是相同的。这些固定的模型是相同的，无论运动是快还是慢，无论哪个自由度对特定任务要求最高的精度都是相同的。在某种程度上，当前BMI的这种局限性反映了关于大脑如何控制不同精确度的运动的基础知识不足。为了推动这一领域的发展，本项目将测试以下假设：1)运动皮质的神经活动根据所需的精度发生特定变化，以及2)更灵活的、调整精度并从多个线性解码器中主动选择的非线性算法将使BMI性能更接近正常人。这一一般性假设将通过两个具体目标进行检验。目的1研究在伸展和抓握任务中，当指令精度发生系统变化时，运动精度的选择性神经编码。目标2将研究使用新的神经信号输入到BMI输出转换来改善BMI性能。这项新颖的BMI实验将专注于使用神经活动的多个维度来提高沿给定自由度的精度。在追求这些具体目标的过程中，我将接受额外的培训，通过开发一个连接运动神经控制、BMI设计和计算神经科学领域的研究计划，来推进我的职业生涯，并从我目前的博士后助理职位过渡到一名独立的终身教职员工。在我的K99年里，我将在Sridei Sarma博士和Robert Jacobs博士的指导下，提高我在计算神经科学方面的培训。与Sarma博士一起，我将重点介绍使用点过程建模对神经活动进行统计建模。与雅各布斯博士一起，我将专注于BMI的混合模型和增益调度的应用。到我的K99岁结束时，我将拥有三个相辅相成的学科的专业知识--运动神经生理学、体重指数和计算神经科学--所有这些都有助于我成功地过渡到独立。

Brain-machine interfaces (BMIs) have progressed rapidly through technological advances that allow recording chronically from large numbers of neurons, more computational power, better training, and better regression algorithms for estimating neural tuning. However, current BMIs rely almost exclusively on fixed, linear models of neural encoding that are the same for all types of movements. These fixed models are the same whether the movements are fast or slow, and the same no matter which degrees of freedom require the most precision for a particular task. In part, this limitation of current BMIs reflects inadequate basic knowledge concerning how the brain controls movement at different levels of precision. To advance the field, the present project will test the hypotheses that 1) specific changes occur in the neural activity of the motor cortex depending on the precision required, and 2) more flexible, non-linear algorithms that adjust precision and actively select from multiple linear decoders will enable BMI performance closer to that of normal humans. This general hypothesis will be tested with two Specific Aims. Aim 1 will examine selective neural encoding of movement precision when the instructed precision is varied systematically in a reach and grasp task. Aim 2 will examine improving BMI performance using novel neural signal input to BMI output transforms. This novel BMI experiment will focus on using multiple dimensions of neural activity to improve precision along a given degree of freedom. In pursuing these Specific Aims, I will receive additional training to advance my career and transition from my current post-doctoral associate position to that of an independent, tenure-track faculty member by developing a research program that bridges the fields of neural control of movement, BMI design, and computational neuroscience. During my K99 years, I will advance my training in computational neuroscience, mentored by Drs. Sridevi Sarma and Robert Jacobs. With Dr. Sarma, I will focus on statistical modeling of neural activity using point process modeling. With Dr. Jacobs, I will focus on applying mixture models and gain-scheduling for BMIs. By the end of my K99 years, I thus will have expertise in three complementary disciplines—motor neurophysiology, BMI, and computational neuroscience—all contributing to the success of my transition to independence.