Prediction of human triglyceride concentration based on quantum genetic-partial least squares algorithm and fluorescence spectroscopy

• Published in: Spectrochimica acta. Part A, Molecular and biomolecular spectroscopy Vol. 251; p. 119461
• Main Authors: Le, De-kang, Zhu, Wei-hua, Zhu, Feng, Yao, Hong-bing, Geng, Ying, Tang, Chun-mei, He, Xiang, Gui, Ming-cheng, Zhang, Lin
• Format: Journal Article
• Language: English
• Published: England Elsevier B.V 15.04.2021
• Subjects: Algorithms; Effective signal intensity; Fluorescence spectra; Humans; Information density; Least-Squares Analysis; Quantum Genetic Algorithm; Spectrometry, Fluorescence; Triglyceride concentration; Triglycerides; Effective signal intensity; Triglyceride concentration; Fluorescence spectra; Information density; Quantum Genetic Algorithm
• ISSN: 1386-1425; 1873-3557
• DOI: 10.1016/j.saa.2021.119461

[Display omitted] •The concept of finite signal intensity was proposed.•Multiple correlations of different types of spectra were studied.•A good prediction model is obtained by combining the QGA with PLS method.•The concept of information density is put forward. It is an important subject with practical significance in modern medical testing about how to obtain various indicators in blood effectively and conveniently. In this essay, the prediction model of triglyceride (TG) concentration was studied based on the fluorescence spectrum of human serum. Firstly, the concept of effective signal intensity was proposed based on the results of wavelet decomposition that the noise signals of spectrum was mainly distributed in the first and second detailed components, and 280 nm was selected as the optimal excitation wavelength for modeling. Secondly, the correlation between fluorescence spectra and triglyceride concentration was studied, which showed that derivative and wavelet decomposition can greatly reduce the multiple correlation of spectrum. Finally, prediction models of triglyceride (TG) concentration were established based on Quantum Genetic Algorithm and Partial Least Squares method, and the result showed that the wavelet decomposition spectral and derivative spectral had better prediction effects because of their lower multiple correlation and advanced resolution, and the Root Mean Square Error reaches to 0.077 mmol/L. In order to obtain the distribution of concentration information in the spectrum, the information density was defined, which indicated that the 3rd layer detailed wavelet decomposition spectrum contains more information of triglyceride concentration. The research results of this essay provide an important reference for the component concentration detection in complex system with multi-component.