Single-model multi-tasks deep learning network for recognition and quantitation of surface-enhanced Raman...

Area of Science:

• Nanoscience and Nanotechnology
• Spectroscopy
• Artificial Intelligence

Background:

• Surface-enhanced Raman scattering (SERS) spectroscopy is crucial for ultrasensitive detection and quantification in nanoscience.
• Analyzing large-scale SERS data and multiple substances presents significant challenges.
• Artificial intelligence (AI) offers advanced tools to revolutionize spectroscopy analysis.

Purpose of the Study:

• To develop a single deep learning model capable of simultaneously performing qualitative recognition and quantitative analysis of SERS spectra.
• To assess the model's accuracy and feasibility for analyzing complex mixtures and real-world samples.

Main Methods:

• A single-model multi-task deep learning network was designed.
• SERS spectra of hypoglycemic drugs (PHE, ROS) and their mixtures were collected across various concentrations (10^-4 M to 10^-8 M).
• Model hyperparameters and loss functions were optimized using the SERS dataset; performance was validated through simulations and analysis in a serum matrix.

Main Results:

• The deep learning model achieved high accuracy rates: 99.0% for qualitative analysis and 98.4% for quantitative analysis.
• The model demonstrated practical feasibility by successfully analyzing PHE and ROS in a complex serum matrix.
• The approach effectively handles multiple substances and large-scale data in SERS analysis.

Conclusions:

• The proposed single-model multi-task deep learning network significantly enhances SERS spectroscopy analysis for recognition and quantification.
• This AI-driven method provides a robust algorithmic and experimental foundation for complex, multi-component SERS applications.
• The study highlights the potential of AI to overcome limitations in traditional SERS data analysis.