Bridging Spectral Gaps: Cross-Device Model Generalization in Blood-Based Infrared Spectroscopy

Area of Science:

• Spectroscopy
• Machine Learning
• Biomedical Engineering

Background:

• Blood infrared spectroscopy is gaining popularity for analysis.
• Machine learning models often struggle with cross-device generalization due to device variations.
• Ensuring model performance across different instruments is crucial for widespread adoption.

Purpose of the Study:

• To develop a method for improving cross-device model generalization in blood infrared spectroscopy.
• To enhance the adaptability of machine learning models to unique device characteristics.
• To validate a novel domain adaptation technique.

Main Methods:

• A straightforward domain adaptation method using data augmentation.
• Incorporating device-specific differences into the augmented training data.
• Experimental validation on two distinct Fourier-Transform Infrared (FTIR) spectroscopy devices.

Main Results:

• The proposed data augmentation method significantly improved prediction accuracy.
• Enhanced model reliability when applied to a different FTIR device.
• Demonstrated effective adaptation to inter-device variations.

Conclusions:

• Data augmentation incorporating device-specific differences is an effective strategy for cross-device generalization in blood infrared spectroscopy.
• The method enhances the practical utility of machine learning models in this field.
• This approach offers a viable solution for deploying spectroscopy models across diverse laboratory settings.