Machine learning approach for rapid and accurate estimation of optical properties using spatial frequency domain...

Area of Science:

• Biomedical Optics
• Medical Imaging
• Machine Learning Applications

Background:

• Quantitative optical properties of tissues are crucial for non-invasive physiological monitoring.
• Current methods for estimating optical properties can be slow and limited in scope.
• Real-time, wide-field mapping of tissue optical properties could significantly advance diagnostic capabilities.

Purpose of the Study:

• To present a novel machine learning-based approach for rapid estimation of tissue optical properties.
• To enable real-time, quantitative mapping of absorption and reduced scattering coefficients.
• To validate the accuracy and speed of the proposed method.

Main Methods:

• Utilized a random forest regression algorithm trained on Monte Carlo photon transport simulation data.
• Estimated optical properties in the spatial frequency domain using diffuse reflectance at two spatial frequencies.
• Applied the trained algorithm to 1-megapixel images for property estimation.

Main Results:

• Achieved high-speed estimation of absorption and reduced scattering coefficients.
• Acquired optical properties over a 1-megapixel image in just 450 milliseconds.
• Demonstrated low estimation errors: 0.556% for absorption and 0.126% for reduced scattering.

Conclusions:

• The machine learning approach enables fast and accurate estimation of tissue optical properties.
• This method facilitates real-time, wide-field quantitative mapping of vital tissue optical parameters.
• The technique holds promise for advancing non-invasive physiological monitoring and medical diagnostics.