RI-Small: Statistical Decoding Models to Improve the Performance of Motor Cortical Brain-Machine Interfaces

该项目的目标是开发统计模型，以准确和有效地解码运动前皮质和运动前皮质中的群体神经元活动。这项研究的重点是运动行为，因为它很容易测量，并且与神经元活动密切相关。运动皮质脑机接口的最新进展表明，研究动物和瘫痪的人类患者能够使用机械臂和计算机光标等外部设备执行基本动作。神经解码向外部设备提供控制命令，通过将大脑信号(例如，一组神经元的峰速)转换为运动状态(例如，手位置、手的运动方向)，在这类接口中起着关键作用。目前的解码模型通常基于神经信号序列是一个平稳过程的强烈假设。然而，这一假设没有考虑到尖峰活动随时间的显著动态变化。此外，这些方法要么专注于解码整个轨迹，要么专注于运动过程中几个“地标”的出现时间。这两种互补策略的有效耦合可以通过更好地利用标志性定义的运动的性质来提高解码性能。该项目将开发计算方法来解决这两个问题。对于非平稳性，研究团队将开发最先进的解码方法的自适应版本，如粒子过滤器和点过程过滤器，可以捕获神经信号中的变化模式并相应地更新模型。为了耦合轨迹解码和时间解码，将从神经活动中识别里程碑式的时间，然后将其合并到运动学模型中。研究小组将使用初级运动皮质、背侧运动前皮质和腹侧运动前皮质多电极阵列的同步记录，这些记录是在行为或视觉运动任务中记录的。改进的解码方法预计将对神经假体产生重大影响。

The goal of this project is to develop statistical models to accurately and efficiently decode population neuronal activity in the motor and premotor cortex. The study focuses on motor behavior as it is easily measured and strongly correlated with neuronal activity. Recent advances in motor cortical brain-machine interfaces have shown that research animals and paralyzed human patients were able to perform rudimentary actions with external devices such as robotic limbs and computer cursors. Neural decoding, which provides control commands to external devices, plays a key role in such interfaces by converting brain signals (e.g., spiking rates of a population of neurons) to kinematic states (e.g., hand position, hand movement direction).Current decoding models are often based on the strong assumption that the neural signal sequence is a stationary process. This assumption, however, does not take into account the significant dynamic variability of spiking activity over time. Moreover, these methods have either focused on decoding the entire trajectory or on the occurrence times of a few "landmarks" during the movement. Effective coupling of these two complementary strategies can be expected to improve the decoding performance by better exploiting the nature of the landmark-defined movement. This project will develop computational methods to address these two issues. For the non-stationarity, the research team will develop adaptive versions of state-of-the-art decoding methods such as particle filters and point process filters that can capture the varying patterns in neural signals and update the model accordingly. To couple trajectory decoding and time decoding, landmark times will be identified from the neural activity, and then incorporated into the kinematic model. The team will use simultaneous recordings from multi-electrode arrays in the primary motor cortex, the dorsal premotor cortex, and the ventral premotor cortex that were recorded during behavior or visuo-motor tasks. Improved decoding methods are expected to have significant impacts on neural prosthetics.