A Systematic Study on Measurement Condition Control and Modeling Methods for Glucose Concentration Prediction Using Near-Infrared Spectroscopy

LIU Wenbo; SUN Di; CHEN Jiayu; YAO Mingfei; LIU Xueying; LIU Jin; HAN Tongshuai; XU Kexin

doi:10.3788/gzxb20265507.0717001

Chinese

您当前的位置：

首页 >

文章列表页 >

A Systematic Study on Measurement Condition Control and Modeling Methods for Glucose Concentration Prediction Using Near-Infrared Spectroscopy

更新时间：2026-06-12

- A Systematic Study on Measurement Condition Control and Modeling Methods for Glucose Concentration Prediction Using Near-Infrared Spectroscopy
- Acta Photonica Sinica Pages: 1-16(2026)
- 作者机构：
  
  1.天津大学精密测试技术及仪器全国重点实验室，天津大学，天津，300072
  2.先阳科技（天津）有限公司，天津，300192
- 作者简介：
- 基金信息：
  
  National Natural Science Foundation of China(81971657)
- DOI：10.3788/gzxb20265507.0717001
  CLC： O433.4
- CSTR：32255.14.gzxb20265507.0717001
- Received：07 March 2026，
  
  Revised：2026-04-15，
  
  Accepted：19 April 2026，
- 稿件说明：
移动端阅览
刘文博，孙迪，陈嘉宇，等. 近红外光谱法糖浓度预测中测量条件控制与建模方法的系统研究［J］.光子学报，2026，55（7）：0717001

LIU Wenbo， SUN Di， CHEN Jiayu， et al. A Systematic Study on Measurement Condition Control and Modeling Methods for Glucose Concentration Prediction Using Near-Infrared Spectroscopy［J］. Acta Photonica Sinica，2026，55（7）：0717001
刘文博，孙迪，陈嘉宇，等. 近红外光谱法糖浓度预测中测量条件控制与建模方法的系统研究［J］.光子学报，2026，55（7）：0717001 DOI： 10.3788/gzxb20265507.0717001. CSTR： 32255.14.gzxb20265507.0717001.

LIU Wenbo， SUN Di， CHEN Jiayu， et al. A Systematic Study on Measurement Condition Control and Modeling Methods for Glucose Concentration Prediction Using Near-Infrared Spectroscopy［J］. Acta Photonica Sinica，2026，55（7）：0717001 DOI： 10.3788/gzxb20265507.0717001. CSTR： 32255.14.gzxb20265507.0717001.

摘要

在基于近红外光谱法的糖浓度预测中，由于糖分子吸收与散射信号微弱，预测精度易受测量条件波动影响。本文结合糖水溶液模拟光谱体系与人体口服葡萄糖耐量试验，系统分析线性与非线性测量条件对光谱特征及模型性能的影响，并比较PLSR与MLP两种建模方法的抗干扰能力，提出“允许变化”与“不允许变化”的测量条件分类策略。仿真结果表明，温度与白噪声是“不允许变化”的，它们呈显著非线性特征，是影响测量精度的关键因素，需将它们分别控制在≤1 ℃和≤0.002 a.u.范围内，预测误差方可稳定在约0.2 mmol/L；而入射角度、光源色温及光源漂移等因素影响近似线性，在较大变化范围内仍可保持较低误差，是“允许变化”的。MLP在处理单一非线性温度干扰时略优于PLSR，但在温度波动较大时，两者误差均明显增大。人体口服葡萄糖耐量试验进一步验证了温控条件下光谱与血糖相关性显著增强，留一受试者法交叉验证预测误差为1.33 mmol/L，优于未温控组（1.63 mmol/L）。可见，合理的温度控制可有效提升预测精度。本研究为近红外光谱法无创血糖测量中的测量条件控制提供了参考。

Abstract

This study aimed to systematically investigate how measurement conditions affect the accuracy and stability of glucose concentration prediction based on near-infrared （NIR） spectroscopy， and to clarify the roles of physical control and modeling methods in improving predictive performance. By combining a simulated glucose-solution spectral system with human oral glucose tolerance test （OGTT） experiments， we examined the effects of linear and nonlinear disturbances on spectral characteristics and model behavior， compared the robustness of partial least squares regression （PLSR） and multilayer perceptron （MLP）， and proposed a classification strategy distinguishing measurement conditions as either “allowable variations” or “non-allowable variations.”A combined simulation-and-human-experiment framework was established. In the simulation study， an aqueous glucose solution was used as a simplified optical system to isolate the effects of controlled disturbances on NIR spectral response and prediction accuracy. Representative factors included liquid temperature variation， white noise， incident angle variation， light-source color temperature， and light-source drift. Their influences were analyzed at both the spectral and modeling levels. In particular， the relationship between disturbance magnitude and unit differential absorbance at a single wavelength was examined to determine whether the induced spectral variation followed an approximately linear or nonlinear pattern. Two modeling approaches were compared： PLSR as a representative linear regression method and MLP as a representative nonlinear learning method. Models were trained to predict glucose concentration under different disturbance scenarios， and performance was evaluated mainly by the root mean square error of prediction （RMSEP）. Based on spectral response patterns and error trends， measurement conditions were classified into “allowable variations，” referring to factors that could vary over a broad range while maintaining low prediction error， and “non-allowable variations，” referring to factors requiring strict control. To verify whether the simulation-based conclusions applied to physiological measurements， human OGTT experiments were also performed. NIR spectra were collected under temperature-controlled and non-temperature-controlled conditions. In the controlled condition， the skin was heated to approximately 36 °C and maintained near this level during measurement. Fingertip blood glucose values were collected intermittently as references. The correlation between spectral signals and blood glucose changes was analyzed under both conditions， with particular attention to features near 1550 nm. Model development included within-group modeling， combined dataset modeling， and leave-one-subject-out cross-validation （LOSOCV）.The simulation results showed that different disturbances had markedly different effects on spectral prediction. Liquid temperature variation and white noise exhibited clear nonlinear characteristics at the single-wavelength level： the relationship between disturbance amplitude and unit differential absorbance was non-proportional， indicating that these factors changed spectral response in a way that could not be adequately represented by simple linear correction. Such nonlinear disturbances altered both the spectral baseline and absorption peak distribution， making it difficult for the models to separate glucose-concentration information from interference. As the amplitudes of temperature fluctuation and noise increased， prediction errors rose substantially. These two factors were therefore identified as “non-allowable variations.” Only when temperature fluctuation was limited to no more than 1 °C and white noise was restricted to no more than 0.002 a.u. could RMSEP remain stably around 0.2 mmol/L. In contrast， incident angle variation， light-source color temperature， and light-source drift showed approximately linear effects within the investigated ranges. The induced spectral changes followed stable patterns， allowing the models to recover glucose-related information with higher robustness. Even when these factors varied over a larger range， prediction errors could still be maintained below about 0.2 mmol/L. These factors were therefore classified as “allowable variations.” Comparison of the two modeling methods showed that MLP had a slight advantage over PLSR when only a limited nonlinear temperature disturbance was present. However， once temperature fluctuation became larger， the errors of both models increased significantly， indicating that nonlinear modeling cannot replace physical control of critical measurement conditions. The human OGTT experiments further supported the simulation findings. Compared with the non-controlled condition， the temperature-controlled condition produced a stronger correlation between NIR spectra and blood glucose changes， with more prominent sensitivity near 1550 nm. Model performance also improved under this condition. In LOSOCV， the MLP model achieved an RMSEP of 1.33 mmol/L in the temperature-controlled group， whereas the corresponding RMSEP in the non-controlled group was 1.63 mmol/L. Similar trends were observed in within-group and combined dataset modeling.This study demonstrates that measurement-condition stability is a key determinant of accurate glucose concentration prediction using NIR spectroscopy. The combined evidence from simulation and human OGTT experiments indicates that nonlinear disturbances are the main factors limiting predictive performance. Temperature variation and white noise were identified as critical “non-allowable variations” because their effects were strongly nonlinear and could not be sufficiently compensated for by either PLSR or MLP. By contrast， incident angle variation， light-source color temperature， and light-source drift behaved approximately linearly and could therefore be tolerated within a wider range as “allowable variations”. Although MLP showed a limited advantage over PLSR under nonlinear interference， model optimization alone was insufficient when disturbance amplitude became large. Human experiments further confirmed that proper temperature control strengthened the spectral correlation with blood glucose and improved prediction under cross-subject validation. Overall， effective physical control of key nonlinear factors is a prerequisite for high-precision prediction.

关键词

Keywords

references

Li T ， Wang Q ， An Y ， et al . Infrared absorption spectroscopy-based non-invasive blood glucose monitoring technology： A comprehensive review ［J］. Biomedical Signal Processing and Control ， 2025 ， 106 ： 107750 .

Davison N B ， Gaffney C J ， Kerns J G ， et al . Recent progress and perspectives on non-invasive glucose sensors ［J］. Diabetology ， 2022 ， 3 （ 1 ）： 56 - 71 .

Liu W ， Han T ， Chen W ， et al . Design key points of high-performance diffuse reflectance optical sensors for non-invasive blood glucose measurement ［J］. Sensors ， 2025 ， 25 （ 4 ）： 998 .

Wang Z ， Zhou L ， Liu T ， et al . Development of non-invasive blood glucose regression based on near-infrared spectroscopy combined with a deep-learning method ［J］. Journal of Physics D： Applied Physics ， 2022 ， 55 （ 21 ）： 215401 .

Yu Y ， Huang J-P ， Zhu J ， et al . An accurate noninvasive blood glucose measurement system using portable near-infrared spectrometer and transfer learning framework ［J］. IEEE Sensors Journal ， 2020 ， 21 （ 3 ）： 3506 - 3519 .

Fu Y ， Huang M ， Chen X . Fingertip capillary dynamic near infrared spectrum （DNIRS） measurement combined with multivariate linear modification algorithm for noninvasive blood glucose monitoring ［J］. Vibrational Spectroscopy ， 2021 ， 113 ： 103223 .

Uwadaira Y ， Ikehata A ， Momose A ， et al . Identification of informative bands in the short-wavelength NIR region for non-invasive blood glucose measurement ［J］. Biomedical Optics Express ， 2016 ， 7 （ 7 ）： 2729 .

Tronstad C ， Elvebakk O ， Staal O M ， et al . Non-invasive prediction of blood glucose trends during hypoglycemia ［J］. Analytica Chimica Acta ， 2019 ， 1052 ： 37 - 48 .

Aloraynan A ， Rassel S ， Xu C ， et al . A single wavelength mid-infrared photoacoustic spectroscopy for noninvasive glucose detection using machine learning ［J］. Biosensors ， 2022 ， 12 （ 3 ）： 166 .

Lundsgaard-Nielsen S M ， Pors A ， Banke S O ， et al . Critical-depth raman spectroscopy enables home-use non-invasive glucose monitoring ［J］. PLOS ONE ， 2018 ， 13 （ 5 ）： e0197134 .

Feng R ， Geng Z ， Li A ， et al . Noninvasive blood glucose monitor via multi-sensor fusion and its clinical evaluation ［J］. Sensors and Actuators B： Chemical ， 2021 ， 332 ： 129445 .

Caduff A ， Dewarrat F ， Talary M ， et al . Non-invasive glucose monitoring in patients with diabetes： A novel system based on impedance spectroscopy ［J］. Biosensors and Bioelectronics ， 2006 ， 22 （ 5 ）： 598 - 604 .

Caduff A ， Talary M S ， Zakharov P . Cutaneous blood perfusion as a perturbing factor for noninvasive glucose monitoring ［J］. Diabetes Technology & Therapeutics ， 2010 ， 12 （ 1 ）： 1 - 9 .

LIU Wenbo ， LIU Jin ， HAN Tongshuai ， et al . Simulation of the effect of dermal thickness on non-invasive blood glucose measurement by near-infrared spectroscopy ［J］. Spectroscopy and Spectral Analysis ， 2023 ， 43 （ 9 ）： 2699 - 2704 .

刘文博，刘瑾，韩同帅，等 . 皮肤真皮层厚度对近红外光谱法无创血糖测量影响的仿真［J］. 光谱学与光谱分析， 2023 ， 43 （ 9 ）： 2699 - 2704 .

Mayrovitz H N . Skin tissue dielectric constant： Time of day and skin depth dependence ［J］. Clinical Physiology and Functional Imaging ， 2025 ， 45 （ 1 ）： e12916 .

Okabe T ， Fujimura T ， Okajima J ， et al . Non-invasive measurement of effective thermal conductivity of human skin with a guard-heated thermistor probe ［J］. International Journal of Heat and Mass Transfer ， 2018 ， 126 ： 625 - 635 .

Han T ， Jin Liu ， Rong Liu ， et al . In vivo near-infrared noninvasive glucose measurement and detection in humans ［J］. Applied Spectroscopy ， 2022 ， 76 （ 9 ）： 1100 - 1111 .

Ge Q ， Han T ， Liu X ， et al . Effects of skin blood flow fluctuations on non-invasive glucose measurement and a feasible blood flow control method ［J］. Sensors ， 2025 ， 25 （ 4 ）： 1162 .

Small G W . Chemometrics and near-infrared spectroscopy： Avoiding the pitfalls ［J］. TrAC Trends in Analytical Chemistry ， 2006 ， 25 （ 11 ）： 1057 - 1066 .

Oshina I ， Spigulis J . Beer–lambert law for optical tissue diagnostics： Current state of the art and the main limitations ［J］. Journal of Biomedical Optics ， 2021 ， 26 （ 10 ）： 100901 .

徐可欣，高峰，赵会娟 . 生物医学光子学［M］. 北京：科学出版社， 2011 .

Wold S ， Sjöström M ， Eriksson L . PLS-regression： a basic tool of chemometrics ［J］. Chemometrics and intelligent laboratory systems ， 2001 ， 58 （ 2 ）： 109 - 130 .

Faber N M ， Song X-H ， Hopke P K . Sample-specific standard error of prediction for partial least squares regression ［J］. TrAC Trends in Analytical Chemistry ， 2003 ， 22 （ 5 ）： 330 - 334 .

Wu Y ， Feng J . Development and application of artificial neural network ［J］. Wireless Personal Communications ， 2018 ， 102 （ 2 ）： 1645 - 1656 .

Yadav J ， Rani A ， Singh V ， et al . Prospects and limitations of non-invasive blood glucose monitoring using near-infrared spectroscopy ［J］. Biomedical Signal Processing and Control ， 2015 ， 18 ： 214 - 227 .

Jensen P S ， Bak J ， Andersson-Engels S . Influence of temperature on water and aqueous glucose absorption spectra in the near- and mid-infrared regions at physiologically relevant temperatures ［J］. Applied Spectroscopy ， 2003 ， 57 （ 1 ）： 28 - 36 .

Amerov A K ， Chen J ， Arnold M A . Molar absorptivities of glucose and other biological molecules in aqueous solutions over the first overtone and combination regions of the near-infrared spectrum ［J］. Applied Spectroscopy ， 2004 ， 58 （ 10 ）： 1195 - 1204 .

LIU Jin ， CAI Zijin ， ZHANG Ziyang ， et al . Evaluation on noninvasive blood components measurement based on photolepthysmography ［J］. Optics and Precision Engineering ， 2016 ， 24 （ 6 ）： 1264 - 1271 .

刘瑾，蔡子晋，张紫杨，等 . 光电容积脉搏波无创测量人体血液成分的评估［J］. 光学精密工程， 2016 ， 24 （ 6 ）： 1264 - 1271 .

Ge Q ， Liu W ， Lu Y ， et al . Skin temperature measurement based on diffuse reflection light at scattering variation independent source-detector separation ［J］. Infrared Physics & Technology ， 2024 ， 139 ： 105295 .

Liu W ， Chen J ， Liu X ， et al . Analysis on direct and additional temperature spectra of skin： A near-infrared diffuse reflectance spectroscopy study with temperature control ［J］. Infrared Physics & Technology ， 2026 ， 152 ： 106187 .

Reminy K ， Hue O ， Antoine-Jonville S . Effect of warm environment on the skin blood flow response to food intake ［J］. International Journal of Hyperthermia ， 2020 ， 37 （ 1 ）， 836 - 842 .

Iredahl F ， Högstedt A ， Henricson J ， et al . Skin glucose metabolism and microvascular blood flow during local insulin delivery and after an oral glucose load ［J］. Microcirculation ， 2016 ， 23 （ 7 ）： 597 - 605 .

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Broadband Silica Waveguide E11-E21/E11-E31 Mode Conversion Switch

Research on Ultra-high Repetition Rate Acousto-optic Frequency Shifting Loop and Spectral Broadening Technology

Fabrication of High Peak Power and High Brightness 808 nm Semiconductor Laser Mini-array

Information Encoding Method Based on Gain-controlled Breathing Dissipative Soliton

Impact of Effective Pixel Count in the Target Beam Reconstruction Plane on Flat-top Beam Shaping and Its Applications

Related Author

No data

Related Institution

No data

AI问答

Postal code：100079
Tel：（010）53879206 Email：tmw@bjxintong.com.cn
Technical support is provided by Beijing Founder electronics co., LTD 京ICP备09082226号-64 京公网安备11010602201714号
It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.

⁰