基于多波长近红外光谱的血红蛋白浓度无创检测技术研究
Research on non-invasive detection technology of hemoglobin concentration based on multi-wavelength near-infrared spectroscopy
彭福来 1陈财 1张宁玲 1王星维 1吕丹阳 1王卫东2
作者信息
- 1. 山东中科先进技术有限公司,济南 250000
- 2. 中国人民解放军总医院 医疗器械研发与临床评价中心,北京 100853
- 折叠
摘要
针对传统的血红蛋白浓度检测需要抽血采样,检测流程复杂且无法连续监测血红蛋白浓度的变化趋势等问题,本研究设计了一种基于多波长近红外光谱的无创血红蛋白浓度检测方法.首先,基于Beer-Lambert定律建立了血红蛋白无创检测数学模型,并依据该模型设计了八波长近红外光电容积脉搏波(photoplethysmography,PPG)信号采集系统;然后,对采集的PPG信号进行降噪和滤除基线漂移等预处理,并根据建立的无创检测模型对特征信息进行提取与选择;最后,基于Stacking算法构建血红蛋白预测回归模型.通过对 249 例临床数据进行实验验证,得到无创检测模型的预测值与参考值的均方根误差为 1.17 g/dL,相关系数为 0.75.实验结果表明,本研究方法可有效实现血红蛋白浓度的无创检测.
Abstract
Aim at the problem that traditional hemoglobin concentration detection requires blood sampling,the detection process is complicated and cannot continuously monitor the variation trend of human hemoglobin concentration,we proposed a non-invasive hemo-globin detection method based on multi-wavelength near-infrared spectroscopy.Firstly,a mathematical model for non-invasive hemo-globin detection was established based on the Beer-Lambert law,and an eight-wavelength near-infrared PPG signal acquisition system was designed based on this model.Then,the collected PPG signals were pre-processed to reduce noise and baseline drift,and the fea-ture information was extracted and selected according to the established non-invasive detection model.Finally,the hemoglobin predic-tion regression model was constructed based on the Stacking algorithm.249 clinical data were used to verify the performance,the root-mean-square error between the predicted value of the non-invasive detection model and the reference value was 1.17 g/dL,and the cor-relation coefficient was 0.75.The experimental results show that this method can achieve non-invasive detection of hemoglobin concentra-tion effectively.
关键词
血红蛋白浓度/无创检测/PPG信号处理/Stacking回归模型Key words
Hemoglobin concentration/Non-invasive detection/Photoplethysmography signal processing/Stacking regression引用本文复制引用
基金项目
山东省自然科学基金资助项目(ZR2020QF024)
山东省自然科学基金资助项目(ZR2021ZD40)
出版年
2024
国家科技期刊平台
NSTL
万方数据