In Vitro Glucose Measurement from NIR and MIR Spectroscopy: Comprehensive Benchmark of Machine Learning and Filtering...

Area of Science:

• Analytical Chemistry
• Spectroscopy
• Machine Learning

Background:

• Accurate glucose quantification is crucial for medical and industrial applications.
• Selecting optimal preprocessing and regression tools for spectroscopic glucose analysis remains a challenge.

Purpose of the Study:

• To comparatively analyze machine learning and preprocessing filter techniques for glucose assay.
• To evaluate the effectiveness of near-infrared (NIR), mid-infrared (MIR), and combined NIR/MIR spectroscopy for glucose prediction.
• To identify the most optimal approach for accurate glucose level prediction.

Main Methods:

• Acquired spectral data from glucose solutions using NIR, MIR, and combined NIR/MIR spectroscopy.
• Applied preprocessing filters: convolutional moving average, Savitzky-Golay, multiplicative scatter correction, and normalization.
• Utilized machine learning algorithms: linear modeling, traditional nonlinear modeling, and artificial neural networks.

Main Results:

• Linear models demonstrated superior predictive accuracy compared to nonlinear models.
• Artificial neural network models showed comparable performance to linear models.
• Convolutional moving average and Savitzky-Golay filters provided the most precise results overall.

Conclusions:

• Appropriate filtering methods significantly enhance predictive accuracy in spectroscopic glucose measurement.
• Linear models combined with specific filters offer a highly effective approach for glucose quantification.
• Findings support the development of advanced glucose monitoring technologies.