Robust estimation of skin physiological parameters from hyperspectral images using Bayesian neural networks

Area of Science:

• Biomedical optics
• Machine learning
• Medical imaging

Background:

• Iterative methods like inverse adding-doubling (IAD) are standard for extracting tissue parameters from hyperspectral images.
• Machine learning offers a faster alternative for direct parameter extraction.

Purpose of the Study:

• To develop a robust Bayesian neural network for predicting physiological parameters from hyperspectral images.
• To create a faster and more accurate method for tissue parameter extraction.

Main Methods:

• A two-component system was developed, modeling spectral-tissue parameter relationships as distributions.
• A neural network was used to approximate the biological tissue model for parameter refinement.
• The model was validated using simulated and in vivo hyperspectral data.

Main Results:

• The proposed Bayesian neural network model achieved a mean absolute error of 0.0141.
• The model demonstrated superior performance compared to existing methods.
• It proved to be a viable, faster alternative to the inverse adding-doubling algorithm.

Conclusions:

• Bayesian neural networks can reliably and accurately extract tissue properties from hyperspectral images.
• This approach enables on-the-fly analysis, significantly speeding up the process.
• The findings support the use of AI in quantitative biomedical imaging.