LA-TReQNet: Improving Multielement Quantification Model for Laser Ablation Inductively Coupled Plasma Mass...

Area of Science:

• Analytical Chemistry
• Geochemistry
• Spectroscopy

Background:

• Laser ablation inductively coupled plasma mass spectrometry (LA-ICP-MS) is a key quantitative technique.
• Conventional calibration methods face practical limitations, including reliance on standards and potential inaccuracies.
• There is a need for more robust and automated calibration strategies in LA-ICP-MS.

Purpose of the Study:

• To introduce LA-TReQNet, a novel end-to-end deep learning framework for fully automated quantitative calibration in LA-ICP-MS.
• To establish a standard-free calibration approach for elemental quantification using deep learning.
• To demonstrate the accuracy and robustness of the proposed method across diverse sample types and data sources.

Main Methods:

• Developed LA-TReQNet, a deep learning framework utilizing a CNN-LSTM architecture.
• Trained the model on a large dataset of 221,364 labeled mass spectra from 5676 samples.
• Implemented an optimized data preprocessing strategy involving power transformer-based standardization and data set grouping.

Main Results:

• LA-TReQNet achieved accurate quantification of 39 elements in independent reference materials, showing robustness to data variations.
• The model demonstrated minimal deviations from certified values: 0.2% ± 5.8% for major elements and -0.9% ± 9.2% for trace elements.
• Deep learning-based quantification matched conventional methods' accuracy while eliminating the need for internal or external standards.

Conclusions:

• LA-TReQNet offers a standard-free calibration approach for LA-ICP-MS, significantly expanding its applicability.
• The framework enhances processing efficiency and result stability by removing reliance on external standards and reducing human variability.
• This deep learning method represents a significant advancement in automated elemental quantification for complex samples.