Study on Non-invasive Blood Glucose Detection Using Near-Infrared Spectroscopy Based on Transfer Learning
10.13241/j.cnki.pmb.2025.13.002
- VernacularTitle:基于迁移学习的近红外光谱非侵入性血糖检测研究
- Author:
Yi-fan LONG
1
;
Le-cheng DING
;
Ze-lin WANG
;
Wei-ze GAO
;
Yong-qian WANG
Author Information
1. 山东大学 山东济南 250000
- Publication Type:Journal Article
- Keywords:
Transfer learning;
Near-infrared spectroscopy;
Non-invasive blood glucose detection;
Machine learning;
Feature selection
- From:
Progress in Modern Biomedicine
2025;25(13):2092-2099
- CountryChina
- Language:Chinese
-
Abstract:
Objective:Near-infrared(NIR)spectroscopy technology faces the problem of insufficient model generalization due to individual differences in non-invasive blood glucose testing,in order to solve this problem,to improve data utilization,and to build predictive models with stronger generalization ability,this study introduces a transfer learning method to study the NIR spectroscopy non-invasive glucose testing.Methods:Migration learning is a machine learning technique that aims to improve task performance in the target domain by transferring knowledge from the source domain to the target domain.In this study,we used community population data as the source domain and student population data as the target domain to improve the performance of the noninvasive glucose detection model on the target domain.In order to verify the effectiveness of migration learning,this study compares the performance of the model before and after migration learning.Results:the model's performance on the noninvasive glucose detection task is significantly improved by the migration learning strategy,and the MAPE and MAE of the migrated model decreases by 52.5460％and 6.0805％,respectively,and the RMSE and MSE decreases by 10.7215％and 12.1135％.Conclusions:This study demonstrates the promising application of transfer learning in the field of non-invasive blood glucose detection,which is expected to realize portable and continuous blood glucose monitoring in the future by migrating the features that are difficult to access in the source domain but are related to blood glucose values to the target domain,which will greatly improve the quality of life of diabetic patients.Advances in noninvasive glucose testing technology will not only reduce patients' pain,but also provide a more convenient means of glucose monitoring,which will provide strong support for diabetes management.
