Machine learning enabled multiple illumination quantitative optoacoustic imaging of blood oxygenation in humans

Thomas Kirchner,Michael Jaeger,Martin Frenz

DOI: https://doi.org/10.48550/arXiv.2202.01482

2022-02-03

Abstract:Optoacoustic (OA) imaging is a promising modality for quantifying blood oxygen saturation (sO$_2$) in various biomedical applications - in diagnosis, monitoring of organ function or even tumor treatment planning. We present an accurate and practically feasible real-time capable method for quantitative imaging of sO$_2$ based on combining multispectral (MS) and multiple illumination (MI) OA imaging with learned spectral decoloring (LSD). For this purpose we developed a hybrid real-time MI MS OA imaging setup with ultrasound (US) imaging capability; we trained gradient boosting machines on MI spectrally colored absorbed energy spectra generated by generic Monte Carlo simulations, and used the trained models to estimate sO$_2$ on real OA measurements. We validated MI-LSD in silico and on in vivo image sequences of radial arteries and accompanying veins of five healthy human volunteers. We compared the performance of the method to prior LSD work and conventional linear unmixing. MI-LSD provided highly accurate results in silico and consistently plausible results in vivo. This preliminary study shows a potentially high applicability of quantitative OA oximetry imaging, using our method.

Medical Physics,Image and Video Processing

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to perform quantitative blood oxygen saturation (sO₂) measurement in human tissues. Specifically, optoacoustic (OA) imaging is a promising biomedical imaging technique that can be used to visualize tissue structure and function, especially through the optical absorption of endogenous or exogenous pigments. However, a long - standing challenge is how to accurately quantify blood oxygen saturation in deep tissues. The main obstacle lies in the ill - posed inverse problem of the light - transport forward problem, that is, the tissue is a scattering medium, and its optical absorption and scattering distributions are unknown, resulting in the unknown distribution of light in the tissue, thus causing the so - called "spectral coloring" phenomenon.

To solve this problem, the paper proposes a method that combines multispectral (MS) and multiple - illumination (MI) optoacoustic imaging and uses a machine - learning method (Learned Spectral Decoloring, LSD) to estimate blood oxygen saturation. The specific steps include:

1. **Develop a real - time multiple - illumination multispectral optoacoustic imaging system**: This system combines ultrasonic imaging capabilities and can be applied to the human body.
2. **Design and perform Monte Carlo simulations**: Generate multi - illumination spectral coloring absorption energy spectra for general human tissues.
3. **Train a gradient - boosting machine model**: Use these simulation data to train the model and perform blood oxygen saturation estimation on actual human optoacoustic measurement data.

The paper demonstrates the high accuracy and robustness of this method in estimating blood oxygen saturation through verification on simulation data and actual human data. Compared with the traditional linear unmixing method, this method shows higher accuracy on both simulation data and actual data.