Rapid platelet dysfunction detection in whole blood samples using machine learning powered micro-clot imaging.

摘要 该项目将优化护理点(POC)血小板力量监测技术，用于临床应用 创伤护理。创伤后死亡和残疾的主要原因与出血和创伤有关。 脑损伤合并颅内出血(ICH)。血小板通过以下途径诱导血栓形成，从而对止血起关键作用 伤口处的粘连、聚集和收缩。创伤后往往会出现血小板功能障碍 加重内出血和脑出血的进展，增加发病率。POC血小板检测还没有被 由于：1)缺乏大队列ED患者检测；2)输血预测准确性差 以及3)延长了处理时间。我们已经制造了一种创新的POC技术来测试血小板功能 直接测量微流体力传感器上的血小板收缩力。我们的PoC测试与 现有的检测方法：1)快速、直接地激活和测量血小板功能；2)创新的机器视觉 具有深刻的机器学习洞察力潜力。然而，这项技术需要在一个 大的主要ED创伤队列，在脑出血后仍未进行测试。我们的飞行员数据显示，血小板的收缩力量 对一系列相关的激活途径和机制很敏感，力在 需要输血的创伤患者。此外，在之前的临床试验中，已经发现血小板输注可以 如果不加区别地使用，是有害的。在这一未得到满足的科学需求的基础上，我们将确定我们的POC 技术可以预测创伤患者的出血性并发症，通知个性化输血 策略。我们最重要的假设是我们的POC血小板力量监测技术是一个有效的指标 创伤和脑出血后出血并发症。目的1：优化血小板力监测光学系统以改进 血小板测力传感器性能。我们假设增加第二个荧光成像通道可以 改进我们目前的血小板测力传感器的性能。目标2：使用机器学习(ML)图像分析 改进对血小板功能障碍的检测和对创伤结局的预测。我们假设以图像为基础 ML模型可以提高测试性能。我们将比较直接测量血小板力的准确性 (AIM 1)与ML增强测量相比，用于检测血小板功能障碍和预测结果。目标3： 验证我们的血小板功能算法用于预测输血需求、死亡率和 严重创伤ED创伤患者中创伤性脑出血的进展。我们 假设血小板力量将是输血需求、死亡率和疾病进展的有力预测因子 我也是。在急诊室中，我们将把我们的算法(包括目标1的原始算法和优化算法)应用于来自 并将预测的输血需求与实际输血(AIM-3a)和 测量测量的血小板力量和脑出血进展之间的关联(AIM-3b)。

Abstract This project will optimize a point-of-care (POC) platelet force monitoring technology for clinical application in trauma care. The leading causes of death and disability after trauma are related to hemorrhage and traumatic brain injury with intracranial hemorrhage (ICH). Platelets are critical to hemostasis by inducing clot formation via adhesion, aggregation, and contraction at wounds. Platelets often become dysfunctional after trauma which worsens internal bleeding and ICH progression and increasing morbidity. POC platelet assays have not been incorporated in practice due to:1) lack of large cohort ED patient testing; 2) poor accuracy in transfusion prediction and 3) extended processing times. We have made an innovative POC technology to test platelet function by directly measuring platelet contractile forces on microfluidic force sensors. Advantages of our POC test vs. existing assays: 1) rapid, direct activation and measures of platelet functions and 2) innovative machine vision with deep potential for machine learning insight. However, this technology needs optimization and validation in a large major ED trauma cohort and remains untested after ICH. Our pilot data suggests platelet contractile forces are sensitive to a range of relevant activation pathways and mechanisms and force is significantly decreased in trauma patients requiring blood transfusion. Further, in prior clinical trials, platelet transfusion has been found to be harmful when used indiscriminately. Building on this unmet scientific need, we will determine if our POC technology is predictive of hemorrhagic complications in trauma patients, informing a personalized transfusion strategy. Our overarching hypothesis is our POC platelet force monitor technology is an efficient indicator of bleeding complications after trauma and ICH. Aim 1: Optimize the platelet force monitor optics to improve platelet force sensor performance. We hypothesize the addition of a second fluorescent imaging channel can improve our current platelet force sensor performance. Aim 2: Use machine learning (ML) image analysis to improve detection of platelet dysfunction and prediction of trauma outcomes. We hypothesize image-based ML models can improve test performance. We will compare the accuracy of direct platelet force measurements (Aim 1) vs. ML-enhancement measurements for detecting platelet dysfunction and predicting outcomes. Aim 3: Validate our platelet function algorithm for predicting blood transfusion needs, mortality, and the progression of traumatic ICH in a prospective cohort of severely-injured ED trauma patients. We hypothesize platelet force will be a powerful predictor of blood transfusion needs, mortality, and progression of ICH. In the ED we will apply our algorithm (both the original and optimized algorithm from Aim 1) to blood from trauma patients and compare the predicted transfusion requirements against actual transfusion (Aim-3a) and measure the association between measured platelet force and ICH progression (Aim-3b).