Learnable real-time inference of molecular composition from diffuse spectroscopy of brain tissue

Area of Science:

• Biomedical Optics
• Medical Physics
• Computational Biology

Background:

• Diffuse optical techniques like broadband near-infrared spectroscopy (bNIRS) and hyperspectral imaging (HSI) offer non-invasive, low-cost, and rapid monitoring of tissue.
• Extracting molecular composition from optical spectra is a key diagnostic capability, but real-time inference methods are lacking, especially for surgical applications.

Purpose of the Study:

• To develop and evaluate a machine learning technique for real-time inference of biochemical composition changes in brain tissue from optical spectra.
• To adapt and assess a learnable methodology based on the Beer-Lambert law for both linear and nonlinear formulations.

Main Methods:

• Modified an existing learnable methodology derived from the Beer-Lambert law.
• Applied the method to data from bNIRS and HSI monitoring of brain tissue.
• Evaluated the technique's performance on linear and nonlinear physical law formulations.

Main Results:

• The proposed method achieves real-time molecular composition inference with accuracy comparable to traditional methods.
• Preliminary results indicate that Beer-Lambert law-based spectral unmixing can differentiate anatomical structures like blood vessels and tumors.
• The technique demonstrates potential for detailed brain anatomy analysis.

Conclusions:

• A data-driven technique for inferring molecular composition changes from diffuse optical spectroscopy of brain tissue has been presented.
• This approach holds promise for enabling critical intra-operative monitoring during neurosurgery.
• The method provides a pathway for advanced, real-time tissue diagnostics using optical spectroscopy.