Peña Edgar, Zhang Simeng, Deyo Steve, Xiao YiZi, Johnson Matthew D
Department of Biomedical Engineering, University of Minnesota, Minneapolis, MN, USA.
J Neural Eng. 2017 Feb;14(1):016014. doi: 10.1088/1741-2552/aa52d1. Epub 2017 Jan 9.
Deep brain stimulation (DBS) therapy relies on both precise neurosurgical targeting and systematic optimization of stimulation settings to achieve beneficial clinical outcomes. One recent advance to improve targeting is the development of DBS arrays (DBSAs) with electrodes segmented both along and around the DBS lead. However, increasing the number of independent electrodes creates the logistical challenge of optimizing stimulation parameters efficiently.
Solving such complex problems with multiple solutions and objectives is well known to occur in biology, in which complex collective behaviors emerge out of swarms of individual organisms engaged in learning through social interactions. Here, we developed a particle swarm optimization (PSO) algorithm to program DBSAs using a swarm of individual particles representing electrode configurations and stimulation amplitudes. Using a finite element model of motor thalamic DBS, we demonstrate how the PSO algorithm can efficiently optimize a multi-objective function that maximizes predictions of axonal activation in regions of interest (ROI, cerebellar-receiving area of motor thalamus), minimizes predictions of axonal activation in regions of avoidance (ROA, somatosensory thalamus), and minimizes power consumption.
The algorithm solved the multi-objective problem by producing a Pareto front. ROI and ROA activation predictions were consistent across swarms (<1% median discrepancy in axon activation). The algorithm was able to accommodate for (1) lead displacement (1 mm) with relatively small ROI (⩽9.2%) and ROA (⩽1%) activation changes, irrespective of shift direction; (2) reduction in maximum per-electrode current (by 50% and 80%) with ROI activation decreasing by 5.6% and 16%, respectively; and (3) disabling electrodes (n = 3 and 12) with ROI activation reduction by 1.8% and 14%, respectively. Additionally, comparison between PSO predictions and multi-compartment axon model simulations showed discrepancies of <1% between approaches.
The PSO algorithm provides a computationally efficient way to program DBS systems especially those with higher electrode counts.
脑深部电刺激（DBS）疗法依赖精确的神经外科靶点定位以及刺激参数的系统优化，以实现有益的临床效果。近期一项改进靶点定位的进展是开发了沿DBS电极及其周围进行分段的DBS阵列（DBSA）。然而，增加独立电极的数量带来了有效优化刺激参数的后勤挑战。
用多种解决方案和目标来解决此类复杂问题在生物学中是常见的，在生物学中，复杂的集体行为源自大量通过社会互动进行学习的个体生物。在此，我们开发了一种粒子群优化（PSO）算法，用代表电极配置和刺激幅度的个体粒子群对DBSA进行编程。使用运动丘脑DBS的有限元模型，我们展示了PSO算法如何能有效地优化一个多目标函数，该函数能使感兴趣区域（ROI，运动丘脑的小脑接收区）的轴突激活预测最大化，使回避区域（ROA，体感丘脑）的轴突激活预测最小化，并使功耗最小化。
该算法通过生成帕累托前沿解决了多目标问题。ROI和ROA激活预测在各粒子群之间是一致的（轴突激活的中位数差异<1%）。该算法能够适应：（1）电极移位（1毫米），ROI（≤9.2%）和ROA（≤1%）激活变化相对较小，且与移位方向无关；（2）最大每电极电流降低（分别降低50%和80%），ROI激活分别降低5.6%和16%；以及（3）禁用电极（n = 3和12），ROI激活分别降低1.8%和14%。此外，PSO预测与多室轴突模型模拟之间的比较表明，两种方法之间的差异<1%。
PSO算法为DBS系统尤其是电极数量较多的系统编程提供了一种计算高效的方法。
