Imaging sub-diffuse optical properties of cancerous and normal skin tissue using machine learning-aided spatial...

Area of Science:

• Biomedical Optics
• Medical Imaging
• Cancer Diagnostics

Background:

• Sub-diffuse optical properties show potential as cancer biomarkers.
• Wide-field heatmaps of these properties can assist in cancerous tissue identification.
• Current sub-diffuse spatial frequency domain imaging (sd-SFDI) methods are too slow for real-time applications.

Purpose of the Study:

• To develop real-time heatmaps of sub-diffuse optical properties using sd-SFDI.
• To report these optical properties for cancerous and normal skin tissue subtypes.

Main Methods:

• Simulated sd-SFDI spectra using a phase function sampling method.
• Trained a machine learning model on simulations and tested it on tissue phantoms.
• Rendered sub-diffuse optical property heatmaps from experimental sd-SFDI images.

Main Results:

• The machine learning model accurately rendered heatmaps from experimental sd-SFDI images in real time.
• Presented heatmaps for cancerous and normal skin tissue subtypes.
• Provided data to inform hypotheses on optical property differences across tissue types.

Conclusions:

• This research significantly accelerates the sd-SFDI process towards real-time capabilities.
• Establishes a foundation for future real-time medical applications, including image-guided surgery.