Machine learning for optical chemical multi-analyte imaging : Why we should dare and why it's not without risks.

Area of Science:

• Biomedical Engineering
• Analytical Chemistry
• Machine Learning Applications

Background:

• Simultaneous sensing of interrelated metabolic analytes like pH and dissolved oxygen (O2) is crucial for understanding complex biological systems.
• Current optical sensing methods face challenges in directly correlating sensor signals to analyte concentrations due to interfering effects.
• Machine learning (ML) offers a promising solution for resolving complex, multidimensional correlations in optical sensing data.

Purpose of the Study:

• To apply machine learning models to fluorescence-based optical chemical sensors for simultaneous multi-analyte imaging in 2D.
• To demonstrate a proof-of-concept for simultaneous imaging of pH and dissolved O2.
• To explore the potential of ML in overcoming challenges in optical chemical sensing.

Main Methods:

• Utilized a fluorescence-based optical chemical sensor array for data acquisition.
• Employed a hyperspectral camera for high-resolution image capture.
• Developed and applied a multi-layered machine learning model, specifically XGBoost, for data analysis and analyte prediction.

Main Results:

• Achieved accurate simultaneous imaging of pH and dissolved O2.
• The ML model predicted dissolved O2 with a mean absolute error < 4.50×10⁻² and pH < 1.96×10⁻¹.
• Reported root mean square errors < 2.12×10⁻¹ for O2 and < 4.42×10⁻¹ for pH.

Conclusions:

• Machine learning, particularly XGBoost, effectively facilitates simultaneous 2D imaging of multiple analytes using optical chemical sensors.
• This approach addresses limitations in correlating sensor signals to analyte concentrations in complex biological samples.
• Highlights the potential of ML for advanced multi-analyte optical sensing while cautioning against potential data biases.