Evaluation of artificial intelligence-controlled CPR to improve vital organ perfusion and survival during prolonged resuscitation

项目总结/摘要 美国每年发生近40万例院外心脏骤停（OHCA）。在 对于需要长时间心肺复苏（CPR）的患者，目前的CPR方法 无法维持足够的血流和氧气输送到重要器官。患者生存率<10% 在可电击节律的患者中约为0%，在不可电击节律的患者中约为0%。当前美国心脏协会 (AHA)关于预防危机和复原的建议遵循"一刀切"的模式。我们的目标是提高生命力 通过"个性化"加压/减压延长CPR期间的器官灌注 使用动态CPR方法进行治疗，该方法在治疗过程中改变按压特征， 心肺复苏术时应考虑胸壁顺应性和血流动力学的时间变化， 增加OHCA后神经系统完整存活率。 在这项拨款申请中，我们正在调查机器学习算法的部署， 用于预测和优化CPR期间的血液动力学。我们将使用最先进的 动态建模结合闭环控制算法， 个性化CPR特征和优化时间血流。我们的初步结果表明， 在临床前心室中部署机器学习预测算法与控制算法配对 纤颤模型可以真实的实时适应压缩和减压深度，从而增加生命力 与标准CPR技术相比，器官血流基于这些结果，我们假设， CPR的按压深度、减压深度、占空比和按压速率的优化将导致 更好的结果。我们提出的研究将：1）确定最有前途的预测算法， CPR血液动力学2）在以下方面识别与该预测算法配对的最佳控制算法： 优化CPR血流动力学和自主循环的恢复3）使用预测和控制配对 与对照组相比， 标准的心肺复苏术在整个过程中，我们将确定非侵入性替代测量， 提供算法，最终目标是继续进行器械开发和人体试验。

Project Summary / Abstract Almost 400,000 cases of out-of-hospital cardiac arrest (OHCA) occur each year in the United States. In patients requiring cardiopulmonary resuscitation (CPR) for prolonged periods, current CPR methods are unable to maintain adequate blood flow and oxygen delivery to the vital organs. Survival is <10% in patients with shockable rhythms and ~0% in those with non-shockable rhythms. Current American Heart Association (AHA) recommendations for CPR follow a “one-size-fits-all” paradigm. Our goal is to improve vital organ perfusion during prolonged CPR by “personalizing” compression/decompression therapy with a dynamic CPR method that changes compression characteristics over the course of CPR after taking into account the temporal changes of chest wall compliance and hemodynamics in order to increase the rate of neurologically intact survival after OHCA. In this grant proposal, we are investigating the deployment of machine learning algorithms incorporated into a mechanical CPR device to predict and optimize hemodynamics during CPR. We will use state-of-the-art dynamical modeling in conjunction with closed-loop control algorithms to individualize CPR characteristics and optimize temporal blood flow. Our preliminary results suggest that deployment of machine learning prediction algorithms paired with control algorithms in a preclinical Ventricular Fibrillation model can adapt compression and decompression depth in real time, resulting in increased vital organ blood flow as compared to standard CPR techniques Based on these results, we hypothesize that optimization of compression depth, decompression depth, duty cycle, and compression rate of CPR will lead to better outcomes. Our proposed research will: 1) identify the most promising algorithm for the prediction of CPR hemodynamics 2) identify the best control algorithm to pair with this prediction algorithm in terms of optimizing CPR hemodynamics and return of spontaneous circulation 3) use the prediction and control pairing to improve 48h neurologically intact survival in a porcine model of ventricular fibrillation, as compared to standard CPR techniques. Throughout this process, we will identify non-invasive alternative measurements to provide to the algorithms with the ultimate goal of proceeding with device development and human trials.