摘要
近红外光谱分析技术在生物医学工程领域具有广阔应用前景。无创且持续性地测量能实时监控人体血糖水平,给糖尿病患者带来极大便利性、提高生存质量、降低糖尿病并发症发生率具有很大的社会效益。无创血糖监测的想法提出较早,但仍然存在预测精度低、预测值与标签值相关性不高等难点,至今没有达到临床要求。近年来,光谱检测技术发展迅猛且机器学习技术在智能信息处理方面具有明显优势,两者结合可以有效提高人体无创血糖医学监测模型的精度和普适性。提出了一种标签敏感度算法(LS),并结合支持向量机方法建立了人体血糖含量预测模型。使用近红外光谱仪采集了4名志愿者食指处动态血液光谱数据(每名志愿者28组数据),并使用多元散射矫正(MSC)方法消除了部分光散射的影响。考虑血糖对不同波长光的吸收有差异,提出了基于血糖浓度标签差的特征波长挑选方法,并构建了标签敏感度支持向量机(LSSVR)预测模型。设计实验,对比该模型与偏最小二乘回归(PLSR)和区分度支持向量机(FSSVR)算法。结果表明,LS算法的最佳特征波长数为32,经特征波长选择后的LSSVR表现最佳,其均方误差降低至0.02 mmol·L^(-1),明显优于全谱段PLSR模型,血糖浓度的预测值与标签值的相关系数提升至99.8%,预测值全部位于可容许误差的克拉克网格A区内。LSSVR模型的优异表现为早日实现血糖的无创监测提供了新思路。
Near-infrared spectroscopy analysis technology has broad application prospects in biomedical engineering.Non-invasive and continuous measurement can monitor the human blood glucose level in real-time,which brings great convenience to diabetes patients,improves the quality of life of patients,and reduces the incidence of complications of diabetes.The idea of non-invasive blood glucose monitoring was put forward earlier,but there are still difficulties,such as low prediction accuracy low correlation between prediction value and label value:up to now,it has not met the clinical requirements.In recent years,spectral detection technology has developed rapidly,and machine learning technology has obvious advantages in intelligent information processing.Combining the two can effectively improve the accuracy and universality of non-invasive blood glucose medical monitoring models.This paper proposes a label sensitivity algorithm(LS),and a prediction model of human blood glucose content is established by combining the support vector machine method.We used a near-infrared spectrometer to collect dynamic blood spectral data at the index finger of four volunteers(28 groups of data for each volunteer)and used the multivariate scattering correction(MSC)method to eliminate the influence of partial light scattering.Considering the difference in the absorption of blood glucose to light of different wavelengths,In this paper,a feature wavelength selection method based on blood glucose concentration label difference is proposed,and a label sensitivity support vector machine(LSSVR)prediction model is constructed Experiments were designed to compare the model with partial least squares regression(PLSR)and discriminant support vector machine(FSSVR,The predicted values are all in the A-region of Clark grid with allowable error.The excellent performance of the LSSVR model provides a new idea for the early realization of non-invasive blood glucose monitoring.
作者
孟琪
赵鹏
宦克为
李野
姜志侠
张瀚文
周林华
MENG Qi;ZHAO Peng;HUAN Ke-wei;LI Ye;JIANG Zhi-xia;ZHANG Han-wen;ZHOU Lin-hua(School of Mathematics and Statistics,Changchun University of Science and Technology,Changchun 130022,China;School of Physics,Changchun University of Science and Technology,Changchun 130022,China;Mathematical Experiment Demonstration Center of Changchun University of Science and Technology,Changchun 130022,China)
出处
《光谱学与光谱分析》
SCIE
EI
CAS
CSCD
北大核心
2024年第3期617-624,共8页
Spectroscopy and Spectral Analysis
基金
吉林省自然科学基金自由探索重点项目(YDZJ202201ZYTS585)
吉林省创新能力建设项目(2022C047-2)
国家自然科学基金项目(11401092)资助。
