An improved system for noninvasive detection of lymphocytes by dynamic spectroscopy

• Published in: Infrared physics & technology 2022-12, Vol.127, p.104423, Article 104423
• Main Authors: Lin, Ling, Jia, Xiuqi, Wang, Dan, Wang, Kang, Wu, Shaohua, Li, Gang
• Format: Article
• Language: English
• Subjects: Lymphocytes; Non-target components; Noninvasive detection; Partial least squares (PLS) modeling; “M+N” theory
• ISSN: 1350-4495; 1879-0275
• DOI: 10.1016/j.infrared.2022.104423

[Display omitted] •In this paper, a system is designed to improve the detection of small blood components by dynamic spectroscopy, which provides a new idea for the spectral analysis of blood.•A partition modeling method considering multiple non-target components is proposed. In the realm of biomedical engineering, improving the prediction accuracy of noninvasive detection of small blood components using spectroscopic approaches has been a challenge. Because blood contains numerous complex components, according to the impact of the M factor in the “M+N” theory, the concentration of non-target components affects the accuracy of spectral detection of target components. This paper investigates an improved system for the noninvasive detection of small blood components based on dynamic spectroscopy and the “M+N” theory. After screening the data quality of the samples collected using dynamic spectroscopy, the remaining 377 samples are subjected to full modeling to determine their predicted concentrations, according to the concentrations of various non-target components like hemoglobin, red blood cells, and platelets. Following that, a complete system for the noninvasive detection of small blood components is constructed by dividing suitable intervals based on non-target components’ predictive values, performing partial least squares (PLS) modeling with lymphocytes as the target component within each interval, and combining them with wavelength screening to improve the target component’s predictive accuracy. When compared to the conventional technique, the correlation coefficient of the prediction set (Rp) increased by about 57.7%, and the root mean square error (RMSEP) decreased by about 76.8%. It has been demonstrated that the system can considerably increase the small blood components’ predicted accuracy of the spectrum analysis, as well as present an idea for solving the problem of spectral detection of small blood components.