Optical-coherence-tomography-based deep-learning scatterer-density estimator using physically accurate noise model.

Area of Science:

• Biomedical Optics
• Medical Imaging
• Machine Learning in Healthcare

Background:

• Optical coherence tomography (OCT) is a crucial imaging modality for visualizing subsurface structures.
• Estimating scatterer density provides insights into tissue microstructure and optical properties.
• Accurate scatterer density estimation is vital for quantitative OCT analysis.

Purpose of the Study:

• To develop and validate a deep-learning-based scatterer density estimator (SDE) for OCT images.
• To improve the accuracy of scatterer density estimation by incorporating spatial noise properties.
• To evaluate the SDE's performance in numerical simulations and experimental settings.

Main Methods:

• A deep-learning model was trained on numerically simulated OCT images with a noise model accounting for shot noise, relative-intensity noise, and non-optical noise.
• The scatterer density estimator (SDE) processes local speckle patterns within OCT images.
• Performance was validated using scattering phantoms and in vitro tumor spheroids.

Main Results:

• The deep-learning-based SDE accurately estimated scatterer densities in OCT images.
• The SDE demonstrated improved estimation accuracy compared to a previous model lacking a comprehensive noise model.
• Validation across numerical and experimental data confirmed the SDE's reliability.

Conclusions:

• The developed deep-learning SDE offers a robust method for quantitative analysis of OCT data.
• Incorporating spatial noise properties in training significantly enhances scatterer density estimation accuracy.
• This advanced SDE has potential applications in various biomedical imaging and diagnostics.