Feng Jingjie, Huang Zhongyi, Zhou Congcong, Ye Xuesong
Biosensor National Special Laboratory, College of Biomedical Engineering and Instrument Science, Zhejiang University, Hangzhou, 310027, People's Republic of China.
State Key Laboratory of CAD & CG, Zhejiang University, Hangzhou, 310027, People's Republic of China.
Australas Phys Eng Sci Med. 2018 Jun;41(2):403-413. doi: 10.1007/s13246-018-0637-8. Epub 2018 Apr 9.
It is widely recognized that pulse transit time (PTT) can track blood pressure (BP) over short periods of time, and hemodynamic covariates such as heart rate, stiffness index may also contribute to BP monitoring. In this paper, we derived a proportional relationship between BP and PPT and proposed an improved method adopting hemodynamic covariates in addition to PTT for continuous BP estimation. We divided 28 subjects from the Multi-parameter Intelligent Monitoring for Intensive Care database into two groups (with/without cardiovascular diseases) and utilized a machine learning strategy based on regularized linear regression (RLR) to construct BP models with different covariates for corresponding groups. RLR was performed for individuals as the initial calibration, while recursive least square algorithm was employed for the re-calibration. The results showed that errors of BP estimation by our method stayed within the Association of Advancement of Medical Instrumentation limits (- 0.98 ± 6.00 mmHg @ SBP, 0.02 ± 4.98 mmHg @ DBP) when the calibration interval extended to 1200-beat cardiac cycles. In comparison with other two representative studies, Chen's method kept accurate (0.32 ± 6.74 mmHg @ SBP, 0.94 ± 5.37 mmHg @ DBP) using a 400-beat calibration interval, while Poon's failed (- 1.97 ± 10.59 mmHg @ SBP, 0.70 ± 4.10 mmHg @ DBP) when using a 200-beat calibration interval. With additional hemodynamic covariates utilized, our method improved the accuracy of PTT-based BP estimation, decreased the calibration frequency and had the potential for better continuous BP estimation.
人们普遍认识到，脉搏传输时间（PTT）可在短时间内追踪血压（BP），而诸如心率、硬度指数等血流动力学协变量也可能有助于血压监测。在本文中，我们推导了血压与PPT之间的比例关系，并提出了一种改进方法，除了PTT之外还采用血流动力学协变量进行连续血压估计。我们将多参数重症监护智能监测数据库中的28名受试者分为两组（有/无心血管疾病），并利用基于正则化线性回归（RLR）的机器学习策略为相应组构建具有不同协变量的血压模型。对个体进行RLR作为初始校准，同时采用递归最小二乘算法进行重新校准。结果表明，当校准间隔延长至1200次心动周期时，我们方法的血压估计误差保持在医学仪器促进协会规定的范围内（收缩压为-0.98±6.00 mmHg，舒张压为0.02±4.98 mmHg）。与其他两项代表性研究相比，Chen的方法在使用400次心跳的校准间隔时保持了较高的准确性（收缩压为0.32±6.74 mmHg，舒张压为0.94±5.37 mmHg），而Poon的方法在使用200次心跳的校准间隔时则出现了误差（收缩压为-1.97±10.59 mmHg，舒张压为0.70±4.10 mmHg）。通过使用额外的血流动力学协变量，我们的方法提高了基于PTT的血压估计的准确性，降低了校准频率，并具有更好地进行连续血压估计的潜力。
