Patient Specific Parameter Optimization of Thalamic Stimulation for Treatment of Epilepsy

)：丘脑前核深部脑刺激（DBS）被临床批准用于治疗癫痫，其癫痫发作频率平均降低40%，但很少有患者实现癫痫发作自由。植入式神经刺激器有许多参数，如刺激幅度、频率和脉冲宽度，这些参数可能被调整以提高疗效。然而，没有系统的过程来指导癫痫医生通过优化。在动物模型中，ANT的刺激显示Papez环的兴奋性几乎立即发生变化，我们假设这是一种可用于优化刺激参数的生物标志物。美敦力的DBS感知系统可以在刺激过程中进行记录，并将数据流式传输到计算机进行进一步分析，这可以用于优化循环。贝叶斯优化（BayesOpt）是一种机器学习算法，广泛用于在获取数据昂贵且计算时间相对便宜的情况下在有限参数范围内进行有效优化。我们使用BayesOpt优化动物模型和临床试验中的刺激设置。在这里，我们建议开发一个优化平台，在这个平台上，医生使用BayesOpt算法的推荐设置来编程刺激设置，以最小化临床中使用Percept从患者丘脑测量到的功率。在探索性临床试验中，提出了开发、测试和验证该方法的三个目标。目标1：开发和测试BayesOpt临床接口与硬件在循环系统。目标2：将BayesOpt应用于20名在临床环境中接受感知系统治疗的癫痫患者，以优化刺激设置，最大限度地减少丘脑活动。目的3：通过对患者进行优化设置和医生选择的设置，验证优化的家庭设置，以测试在优化设置中癫痫发作频率或功率谱密度是否显着降低。本临床试验的结果将是建立安全性和可行性的优化和验证。如果成功，该研究将用于II期疗效试验。这项工作的更广泛的影响是，这个平台可以用来调整感知系统，基于不同的生物标志物，在其他疾病，如帕金森病，疼痛和抑郁症。

): Deep brain stimulation (DBS) of the anterior nucleus of the thalamus (ANT) is clinically approved for treatment of epilepsy resulting in an average decrease in seizure frequency of 40%, but few patients achieve seizure freedom. Implantable neural stimulators have many parameters, such as stimulation amplitude, frequency and pulse width, which could potentially be tuned to improve efficacy. However, there is no systematic process to guide epileptologists through optimization. Stimulation of ANT in animal models has shown almost immediate changes in excitability in the loop of Papez, which we hypothesize is a biomarker that could be used to optimize stimulation parameters. Medtronic’s DBS Percept system allows for recording during stimulation and streaming the data to a computer for further analysis, which can be used in an optimization loop. Bayesian optimization (BayesOpt) is a machine learning algorithm that is widely used for efficient optimization over a bounded parameter range when acquiring data is expensive and computational time is relatively cheap. We have used BayesOpt for optimizing stimulation settings in animal models and clinical trials. Here we propose to develop an optimization platform where stimulation settings are programmed by a physician using recommended settings from a BayesOpt algorithm to minimize power measured from the patient’s thalamus in the clinic using Percept. Three aims are proposed to develop, test, and validate this approach in an exploratory clinical trial. Aim 1: Develop and test BayesOpt clinical interface with hardware in the loop system. Aim 2: Apply BayesOpt to 20 epilepsy patients treated with the Percept system in a clinical setting to optimize stimulation settings to minimize thalamic activity. Aim 3: Validate optimized settings at home by programming patients with optimized setting and their physician selected setting to test if seizure frequency or power spectral density is significantly lower in the optimized setting. The outcome of this clinical trial will be to establish safety and feasibility of optimization and validation. If successful, this study will be used to power a phase II efficacy trial. The broader impact of this work is that this platform could be used to tune the Percept system, based on different biomarkers, in other diseases, such as Parkinson’s disease, pain, and depression.