Bimodal spectroscopy integrating multi-wavelength time-resolved photoacoustic spectroscopy and near-infrared spectroscopy with deep learning for quantitative detection of serum biochemical indicators
- 1Key Laboratory of Advanced Electronic Materials and Devices of Jiangxi Province, Jiangxi Science and Technology Normal University, Nanchang, Jiangxi 330038, China
- 2Jiangxi Province Key Laboratory of Immunology and Inflammation, Jiangxi Provincial Clinical Research Center for Laboratory Medicine, Department of Clinical Laboratory, The Second Affiliated Hospital, Jiangxi Medical College, Nanchang University, Nanchang, Jiangxi 330038, China
Abstract
To achieve accurate and high-efficient quantitative detection of three serum biochemical indicators (SBIs), namely glucose (GLU), triglycerides (TG), and total cholesterol (TC), this study proposes a bimodal spectroscopy method integrating multi-wavelength time-resolved photoacoustic spectroscopy (TR-PAS) with near-infrared spectroscopy (NIRS), combined with a deep learning (DL) approach. A total of 35 characteristic wavelengths (12 for GLU, 13 for TG, and 10 for TC) are determined from the energy-corrected photoacoustic (PA) peak-to-peak value spectra of biochemical standard solutions by integrating three wavelength selection methods. NIR spectra and time-resolved PA signals at the selected wavelengths are experimentally collected from 1084 serum samples. Following data augmentation for the bimodal data and data preprocessing for NIR spectra, a dual-branch DL network, namely DB-CNN-LSTM-MAM model, was established to quantitatively predict GLU, TG, and TC, respectively. Through optimization of the model structure and parameters, the root mean square error of prediction (RMSEP) and coefficient of determination (R p ²) for the testing set were as follows: GLU: 0.9653 mmol/L, 0.9296; TG: 0.3956 mmol/L, 0.9575; TC: 0.4443 mmol/L, 0.9290. The generalization of the established DL model is verified by using 10-fold cross validation, and the external independent samples. The quantitiative performances of three SBIs are compared with other DL models. To validate the effectiveness of the proposed method, the comparisons are also performed between the bimodal spectroscopy and unimodal spectroscopy, as well as between the multi-wavelength TR-PAS–NIRS and the single-wavelength TR-PAS–NIRS. The study results indicate that the proposed method provides an accurate and highly efficient candidate solution for the quantitative detection of SBIs of ex vivo serum.
Related
- Evaluating and choosing experimental parameters for shifted excitation Raman difference spectroscopy: a quantitative study
- Fecal Raman Spectroscopy Complements Microbiome and Metabolomic Analyses to Track Growth-Stage Biochemical Changes in Sika Deer
- NA-CDQuant: Quantitative spectral analysis of nucleic acid structure by circular dichroism spectroscopy
- Machine Learning Integrating Quantitative MRI and Clinical Features Predicts Loss of Ambulation in Duchenne Muscular Dystrophy.
- Hyaluronic Acid-Based Nanoplatforms for Photoacoustic Monitoring of Phototherapy
BibTeX
@article{ren2026bimodal,
title = {{Bimodal spectroscopy integrating multi-wavelength time-resolved photoacoustic spectroscopy and near-infrared spectroscopy with deep learning for quantitative detection of serum biochemical indicators}},
author = {Zhong Ren and Chaojun Chen and Gaoqiang Liang and Haibin Zhang and Weinan Shi and Jia Zhang and Guohui Xiao and Xiaoyu Zhu and Wenyan Nie},
journal = {Photoacoustics},
year = {2026},
month = oct,
doi = {10.1016/j.pacs.2026.100858},
url = {https://doi.org/10.1016/j.pacs.2026.100858},
note = {CC BY},
}
RIS
TY - JOUR
TI - Bimodal spectroscopy integrating multi-wavelength time-resolved photoacoustic spectroscopy and near-infrared spectroscopy with deep learning for quantitative detection of serum biochemical indicators
AU - Zhong Ren
AU - Chaojun Chen
AU - Gaoqiang Liang
AU - Haibin Zhang
AU - Weinan Shi
AU - Jia Zhang
AU - Guohui Xiao
AU - Xiaoyu Zhu
AU - Wenyan Nie
PY - 2026
DA - 2026/10/29
JO - Photoacoustics
DO - 10.1016/j.pacs.2026.100858
UR - https://doi.org/10.1016/j.pacs.2026.100858
AB - To achieve accurate and high-efficient quantitative detection of three serum biochemical indicators (SBIs), namely glucose (GLU), triglycerides (TG), and total cholesterol (TC), this study proposes a bimodal spectroscopy method integrating multi-wavelength time-resolved photoacoustic spectroscopy (TR-PAS) with near-infrared spectroscopy (NIRS), combined with a deep learning (DL) approach. A total of 35 characteristic wavelengths (12 for GLU, 13 for TG, and 10 for TC) are determined from the energy-corrected photoacoustic (PA) peak-to-peak value spectra of biochemical standard solutions by integrating three wavelength selection methods. NIR spectra and time-resolved PA signals at the selected wavelengths are experimentally collected from 1084 serum samples. Following data augmentation for the bimodal data and data preprocessing for NIR spectra, a dual-branch DL network, namely DB-CNN-LSTM-MAM model, was established to quantitatively predict GLU, TG, and TC, respectively. Through optimization of the model structure and parameters, the root mean square error of prediction (RMSEP) and coefficient of determination (R p ²) for the testing set were as follows: GLU: 0.9653 mmol/L, 0.9296; TG: 0.3956 mmol/L, 0.9575; TC: 0.4443 mmol/L, 0.9290. The generalization of the established DL model is verified by using 10-fold cross validation, and the external independent samples. The quantitiative performances of three SBIs are compared with other DL models. To validate the effectiveness of the proposed method, the comparisons are also performed between the bimodal spectroscopy and unimodal spectroscopy, as well as between the multi-wavelength TR-PAS–NIRS and the single-wavelength TR-PAS–NIRS. The study results indicate that the proposed method provides an accurate and highly efficient candidate solution for the quantitative detection of SBIs of ex vivo serum.
SN - 2213-5979
ER -