Multiple illumination learned spectral decoloring for quantitative optoacoustic oximetry imaging

Area of Science:

• Optoacoustic (OA) imaging
• Biomedical optics
• Quantitative imaging

Background:

• Accurate blood oxygen saturation (sO2) measurement is crucial for various biomedical applications.
• Optoacoustic (OA) imaging offers a promising non-invasive approach for quantitative sO2 monitoring.
• Existing OA imaging methods face challenges in achieving precise and real-time sO2 quantification.

Purpose of the Study:

• To develop and validate a novel method for accurate and real-time quantification of local sO2 using OA imaging.
• To combine multiple illumination (MI) sensing with learned spectral decoloring (LSD) for enhanced sO2 estimation.
• To assess the performance of the MI-LSD method using a reliable phantom model.

Main Methods:

• The study combined multiple illumination (MI) sensing with learned spectral decoloring (LSD).
• Feedforward neural networks and random forests were trained on Monte Carlo simulations of spectral data.
• The trained models were applied to real OA measurements and validated on a copper and nickel sulfate solution phantom model.

Main Results:

• The MI-LSD method achieved consistently high estimation accuracy, with median absolute errors ranging from 2.5 to 4.5 percentage points.
• MI-LSD demonstrated a significant reduction in outliers compared to LSD alone.
• Random forest regressors showed superior performance over previously reported neural network approaches.

Conclusions:

• The developed random forest-based MI-LSD method is a promising advancement for accurate quantitative OA oximetry imaging.
• This approach offers improved accuracy and reliability for non-invasive blood oxygen saturation monitoring.
• The findings support the potential of MI-LSD for widespread biomedical applications requiring precise sO2 quantification.