Related Experiment Video
Updated: Jul 10, 2026
Automated Analysis of C. elegans Fluorescence Images using SegElegans
Published on: October 10, 2025
Physics-guided deep learning for improved stability and interpretability in EEM fluorescence decomposition
Beibei Xie1, Huikang Li2, Yan Zou2
1School of Artificial Intelligence (School of Software), Yanshan University, Qinhuangdao, Hebei 066004, China; Hebei Key Laboratory of Computer Virtual Technology and System Integration, Qinhuangdao, Hebei 066004, China.
A new framework for analyzing dissolved organic matter (DOM) using excitation-emission matrix (EEM) fluorescence spectroscopy offers stable and interpretable results. This method uses synthetic data for calibration, improving decomposition accuracy even with limited real-world measurements.
Area of Science:
- Environmental Chemistry
- Spectroscopy
- Data Analysis
Background:
- Three-dimensional excitation-emission matrix (EEM) fluorescence spectroscopy is crucial for characterizing dissolved organic matter (DOM).
- Traditional methods like parallel factor analysis (PARAFAC) face challenges with quality control and preprocessing.
- Machine learning approaches are limited by small and sparse EEM datasets.
Purpose of the Study:
- To introduce a unified, physically consistent framework for EEM decomposition.
- To address limitations of existing methods by decoupling spectral shape and intensity.
- To enable stable inverse mapping for EEM data analysis.
Main Methods:
- Developed a dual normalization technique to stabilize EEM decomposition.
- Utilized synthetic mixtures from OpenFluor reference spectra for supervised training.
- Evaluated the inverse solver on experimentally measured EEM datasets.
Main Results:
- Achieved accurate reconstructions on synthetic data with mean squared error ~10-6.
- Demonstrated stable performance on real measurements with mean absolute reconstruction error of 10-3-10-2.
- Recovered component spectra closely matched reference features with a mean maximum similarity of ~0.94.
Conclusions:
- Physically constrained synthetic calibration enables stable and interpretable EEM decomposition with limited real data.
- The proposed framework offers a robust alternative to conventional PARAFAC and data-driven methods.
- The approach shows potential for extension to other linearly mixed spectroscopic techniques.
