College of Information Science and Technology, Beijing University of Chemical Technology, China.
College of Electrical Engineering and Automation, Shandong University of Science and Technology, 579# QianWanGang Road, Qingdao 266590, China; College of Information Science and Technology, Beijing University of Chemical Technology, China.
Comput Methods Programs Biomed. 2018 Mar;156:61-71. doi: 10.1016/j.cmpb.2017.12.019. Epub 2017 Dec 22.
Intelligent status monitoring for critically ill patients can help medical stuff quickly discover and assess the changes of disease and then make appropriate treatment strategy. However, general-type monitoring model now widely used is difficult to adapt the changes of intensive care unit (ICU) patients' status due to its fixed pattern, and a more robust, efficient and fast monitoring model should be developed to the individual.
A data-driven learning approach combining locally weighted projection regression (LWPR) and principal component analysis (PCA) is firstly proposed and applied to monitor the nonlinear process of patients' health status in ICU. LWPR is used to approximate the complex nonlinear process with local linear models, in which PCA could be further applied to status monitoring, and finally a global weighted statistic will be acquired for detecting the possible abnormalities. Moreover, some improved versions are developed, such as LWPR-MPCA and LWPR-JPCA, which also have superior performance.
Eighteen subjects were selected from the Physiobank's Multi-parameter Intelligent Monitoring for Intensive Care II (MIMIC II) database, and two vital signs of each subject were chosen for online monitoring. The proposed method was compared with several existing methods including traditional PCA, Partial least squares (PLS), just in time learning combined with modified PCA (L-PCA), and Kernel PCA (KPCA). The experimental results demonstrated that the mean fault detection rate (FDR) of PCA can be improved by 41.7% after adding LWPR. The mean FDR of LWPR-MPCA was increased by 8.3%, compared with the latest reported method L-PCA. Meanwhile, LWPR spent less training time than others, especially KPCA.
LWPR is first introduced into ICU patients monitoring and achieves the best monitoring performance including adaptability to changes in patient status, sensitivity for abnormality detection as well as its fast learning speed and low computational complexity. The algorithm is an excellent approach to establishing a personalized model for patients, which is the mainstream direction of modern medicine in the following development, as well as improving the global monitoring performance.
对危重症患者进行智能状态监测，可以帮助医护人员快速发现和评估疾病变化，从而制定出恰当的治疗策略。然而，目前广泛使用的通用型监测模型由于模式固定，难以适应重症监护病房（ICU）患者状态的变化，因此需要开发更强大、高效和快速的针对个体的监测模型。
本文首次提出了一种基于数据驱动的学习方法，结合局部加权投影回归（LWPR）和主成分分析（PCA），用于监测 ICU 患者健康状态的非线性过程。LWPR 用于用局部线性模型逼近复杂的非线性过程，其中可以进一步应用 PCA 进行状态监测，最后获得全局加权统计量以检测可能的异常。此外，还开发了一些改进版本，如 LWPR-MPCA 和 LWPR-JPCA，它们也具有优异的性能。
从 Physiobank 的多参数智能监护重症监护 II （MIMIC II）数据库中选择了 18 个受试者，每个受试者选择了两个生命体征进行在线监测。将所提出的方法与几种现有方法进行了比较，包括传统 PCA、偏最小二乘法（PLS）、即时学习与改进 PCA（L-PCA）和核 PCA（KPCA）。实验结果表明，在添加 LWPR 后，PCA 的平均故障检测率（FDR）可以提高 41.7%。与最新报道的方法 L-PCA 相比，LWPR-MPCA 的平均 FDR 提高了 8.3%。同时，LWPR 的训练时间比其他方法都要少，尤其是 KPCA。
本文首次将 LWPR 引入 ICU 患者监测中，并实现了最佳的监测性能，包括对患者状态变化的适应性、对异常检测的敏感性以及其快速学习速度和低计算复杂度。该算法是为患者建立个性化模型的优秀方法，是现代医学在未来发展中的主流方向，同时也提高了整体监测性能。
Zotero 插件